feat: feature, package, and custom regeneration inputs

[Verdenroz]

What changed

The action ran the CLI with no --features, so a bench target that declares required-features could not be gated at all, and a changelog built without a crate's full feature set silently drops every feature-gated public item from its API diff. A crate with no bench target still ships an API the changelog should diff, and derived files outside the built-in regeneration (an sdk gen, a rendered report) had no way to land through the bot. This adds inputs for each.

• Added features (passed to the gate, the baseline measurement, and spec generation) and changelog-features (passed to report changelog, defaulting to features). They can differ on purpose: gating under a heavy feature set can be too slow while the changelog's API surface needs all of it.
• Added changelog-packages (defaulting to packages) so the changelog can diff crates the gate does not measure.
• Added regen-run and regen-paths: a shell snippet run on the default branch after the built-in regeneration, with $SOOTHFAST naming the installed CLI, whose changes under the given pathspecs land in the same bot pull request. regen-run without regen-paths fails the step rather than discarding what it changed.
• Documented the five inputs in docs/ci.md.

No CLI change: every flag the scripts now pass exists in cargo-soothfast 0.3.0.

Why

The one-step promise only holds if the step can run a repository's benches as they are declared. required-features on a bench target and crates that ship an API without a bench are ordinary Cargo layouts, and sdk gen is this project's own feature. Each was a workaround an adopter would otherwise need, which the mechanism-design rule in CLAUDE.md says to avoid.

How was this tested

• shellcheck -x on all eight scripts: clean. zizmor 1.29.0 on action.yml: no findings.
• An empty input passes no flag: the scripts build the --features arguments as arrays, verified under set -u.
• make check and make gate not run; no Rust changed.

Checklist

• make check passes (fmt, clippy -D warnings, cargo test --workspace)
• make gate BASE=master passes, or any intentional cost change is explained above
• Tests added/updated for new behavior (not required for docs/CI-only changes)
• Public API items have /// doc comments
• No new dependency, or its justification is included above (see Dependency Policy)
• Docs (README.md, docs/, soothfast:bind/soothfast:claim markers) updated if behavior changed

[soothfast-bot]

soothfast gate

gate: measuring merge-base of origin/master in worktree (interleaved rounds)
gate: bench binaries identical (code and data match) — no measurable change possible
gate: gating backend = walltime
gate: build=08f03a noise_floor=0.14% thresholds: instructions +5% ir +5% walltime +10.0% alloc/size +5% polls/wakes +5%
ok    soothfast_spec::bench_openapi_diff walltime_median_ns 6378510.0 -> 6378510.0 (+0.0%)
ok    soothfast_spec::bench_openapi_diff allocs 77722 -> 77722 (allowed <= 81608)
ok    soothfast_spec::bench_openapi_diff alloc_bytes 10815487 -> 10815487 (allowed <= 11356261)
ok    soothfast_spec::bench_openapi_document walltime_median_ns 2006512.0 -> 2006512.0 (+0.0%)
ok    soothfast_spec::bench_openapi_document allocs 30152 -> 30152 (allowed <= 31659)
ok    soothfast_spec::bench_openapi_document alloc_bytes 4715065 -> 4715065 (allowed <= 4950818)
ok    soothfast_spec::bench_serialize_yaml walltime_median_ns 6600786.0 -> 6600786.0 (+0.0%)
ok    soothfast_spec::bench_serialize_yaml allocs 74277 -> 74277 (allowed <= 77990)
ok    soothfast_spec::bench_serialize_yaml alloc_bytes 8533615 -> 8533615 (allowed <= 8960295)
ok    soothfast_spec::bench_openapi_diff assert complexity: claimed O(n); growth drift x1.73 over sizes [16, 64, 256] (limit x2.5)
ok    soothfast_spec::bench_openapi_document assert complexity: claimed O(n); growth drift x1.50 over sizes [16, 64, 256] (limit x2.5)
ok    soothfast_spec::bench_serialize_yaml assert complexity: claimed O(n); growth drift x1.23 over sizes [16, 64, 256] (limit x2.5)
gate: passed (3 item(s))