ci: collapse the two bot regeneration commits into one

[Verdenroz]

Description

Every human merge to master lands two soothfast-bot commits and runs three Docker build-and-deploy cycles. Two workflows each regenerate half the derived artifacts and each open their own bot PR, and nothing stops those bot commits from rebuilding all three images. The deploy-docs job already runs last and already builds the soothfast-routes bench target, so it can produce the changelog and the specs as well. Moving that work there and skipping the rebuild for the bot's own commit leaves one bot PR and one build cycle per merge.

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that causes existing functionality to change)
• Documentation update
• Refactoring (no functional changes)
• Performance improvement

Changes

• Moved the spec gen, sdk gen, and MCP spec gen regeneration out of the gate job's regen-run in soothfast.yml into a step in deploy.yml's deploy-docs, placed after the endpoint reference pages so it reuses the bench binary already built there.
• Added a report changelog step to deploy-docs, before the token mint so every step that compiles the tree still runs first. It resolves the latest tag with git describe and passes it as --against-ref.
• Added fetch-tags: true to the deploy-docs checkout. actions/checkout passes --no-tags by default even at fetch-depth: 0, and without a tag the changelog step would replace the real Unreleased section with an initial-surface dump.
• Extended deploy-docs's land.sh PATHS to the union of both landings: CHANGELOG.md, .soothfast/trend.jsonl, llms.txt, docs/library, README.md, server/openapi.yaml, server/asyncapi.yaml, sdks, finance-query-mcp/mcp-tools.json.
• Set changelog: false on the Verdenroz/soothfast action in the gate job and dropped its regen-run, regen-paths, and changelog-features inputs. With no regen inputs the action's regen step collects no pathspecs and reports changed=false, so it mints no token and opens no pull request.
• Set save-if: false on the gate job's rust-cache. It now compiles nothing on a master push and would finish first, saving a near-empty target under the bench-gate key it shares with baseline.
• Guarded build-and-push and deploy-docs with if: github.actor != 'soothfast-bot[bot]'. deploy has needs: build-and-push, so the VPS deploy skips along with it.
• Updated the header comment and the gate job comment in soothfast.yml, which described the regeneration that job no longer does.

The guard matches the soothfast bot by name rather than testing for any [bot] actor. Dependabot's base-image digest bumps for debian and rust must still rebuild the images.

Breaking Changes

None. No library, server, or MCP code changed.

Testing

• Unit tests added/updated
• Integration tests added/updated
• Manual testing performed
• All tests pass (cargo test)

Verified statically: both workflows parse under yaml.safe_load, prek run --files on both is clean, and zizmor --no-online-audits over both reports no findings. actionlint is not installed on this machine, so it did not run. No Rust changed, so no cargo gate applies.

The github.actor guard was checked against real run payloads before it was written. Runs 34423002373 and 34042536402, the two soothfast-bot commits, report actor.login as soothfast-bot[bot]. Run 33565318087, a dependabot base-image bump, reports actor.login as Verdenroz with head_commit.author as dependabot[bot], which is why the guard cannot key off the commit author.

Three behaviors cannot be verified before merge: that exactly one bot PR opens per human merge, that the rebuild is skipped for it, and that the changelog regenerates with its entries intact. The first merge to master after this is the test. What to watch for:

• One bot/derived-artifacts PR and no bot/soothfast-update PR.
• Build and Deploy skipped on the bot's own push.
• CHANGELOG.md still headed ## Unreleased (draft vs v3.0.0) with its entry lists intact, not ## Unreleased (initial public surface).

That last one is the rollback trigger. If the heading changes, revert this rather than hand-editing the file.

Documentation

• Code documentation updated (doc comments)
• README updated (if needed)
• CHANGELOG.md updated (if releasing)
• Examples updated (if API changed)

Checklist

• Code compiles without warnings (cargo check)
• Tests pass (cargo test)
• Code formatted (cargo fmt)
• Clippy checks pass (cargo clippy)
• Documentation builds (cargo doc --no-deps)
• Commit messages follow Conventional Commits

Related Issues

Closes #479

[soothfast-bot]

soothfast gate

finance-query

ok    finance_query::ticker_quote_then_cached allocs 4 -> 4 (allowed <= 4)
ok    finance_query::ticker_quote_then_cached alloc_bytes 43864 -> 43864 (allowed <= 46057)
ok    finance_query::tickers_batch_quote_then_cached walltime_median_ns 10450.6 -> 10450.6 (+0.0%)
ok    finance_query::tickers_batch_quote_then_cached allocs 16 -> 16 (allowed <= 16)
ok    finance_query::tickers_batch_quote_then_cached alloc_bytes 134810 -> 134810 (allowed <= 141550)
ok    finance_query::translate_dictionary walltime_median_ns 1958.4 -> 1958.4 (+0.0%)
ok    finance_query::translate_dictionary allocs 49 -> 49 (allowed <= 51)
ok    finance_query::translate_dictionary alloc_bytes 2880 -> 2880 (allowed <= 3024)
ok    finance_query::translation_set_backend walltime_median_ns 17.8 -> 17.8 (+0.0%)
ok    finance_query::translation_set_backend allocs 1 -> 1 (allowed <= 1)
ok    finance_query::translation_set_backend alloc_bytes 16 -> 16 (allowed <= 16)
ok    finance_query::translation_translate walltime_median_ns 3674.2 -> 3674.2 (+0.0%)
ok    finance_query::translation_translate allocs 79 -> 79 (allowed <= 82)
ok    finance_query::translation_translate alloc_bytes 6298 -> 6298 (allowed <= 6612)
ok    finance_query::translation_translate_with walltime_median_ns 3541.7 -> 3541.7 (+0.0%)
ok    finance_query::translation_translate_with allocs 77 -> 77 (allowed <= 80)
ok    finance_query::translation_translate_with alloc_bytes 6294 -> 6294 (allowed <= 6608)
ok    finance_query::bt_base_to_htf_index assert complexity: claimed O(n); growth drift x1.28 over sizes [1024, 4096, 16384] (limit x2.5)
ok    finance_query::bt_resample assert complexity: claimed O(n); growth drift x0.96 over sizes [1024, 4096, 16384] (limit x2.5)
ok    finance_query::bt_sma_crossover assert complexity: claimed O(n); growth drift x1.03 over sizes [1024, 4096, 16384] (limit x2.5)
ok    finance_query::capability_name assert alloc: allocs 0 <= 0
ok    finance_query::de_quote assert p99: p99 1101740ns <= 3000000ns
ok    finance_query::dispatch_select assert alloc: allocs 0 <= 0
ok    finance_query::ind_patterns assert complexity: claimed O(n); growth drift x1.11 over sizes [1024, 4096, 16384] (limit x2.5)
ok    finance_query::ind_sma assert alloc: allocs 1 <= 1
ok    finance_query::ind_sma assert p99: p99 432832ns <= 1000000ns
ok    finance_query::ind_sma assert complexity: claimed O(n); growth drift x1.01 over sizes [1000, 10000, 100000] (limit x2.5)
ok    finance_query::risk_beta assert alloc: allocs 0 <= 0
ok    finance_query::risk_beta assert complexity: claimed O(n); growth drift x1.03 over sizes [1024, 4096, 16384] (limit x2.5)
ok    finance_query::risk_calmar assert alloc: allocs 0 <= 0
ok    finance_query::risk_historical_var assert alloc: allocs 2 <= 2
ok    finance_query::risk_historical_var assert complexity: claimed O(n log n); growth drift x1.05 over sizes [1000, 10000, 100000] (limit x2.5)
ok    finance_query::risk_max_drawdown assert alloc: allocs 1 <= 1
ok    finance_query::risk_parametric_var assert alloc: allocs 0 <= 0
ok    finance_query::risk_parametric_var assert complexity: claimed O(n); growth drift x1.01 over sizes [1024, 4096, 16384] (limit x2.5)
ok    finance_query::risk_sharpe assert alloc: allocs 0 <= 0
ok    finance_query::risk_sharpe assert complexity: claimed O(n); growth drift x1.01 over sizes [1024, 4096, 16384] (limit x2.5)
ok    finance_query::risk_sortino assert alloc: allocs 0 <= 0
ok    finance_query::risk_sortino assert complexity: claimed O(n); growth drift x1.02 over sizes [1024, 4096, 16384] (limit x2.5)
gate: passed (180 item(s))