feat(model): tan model edge-AI lifecycle surface (check/zoo/add/prep/run/ab) — consolidated

[alpCaner]

Summary

Consolidated tan model edge-AI lifecycle surface — the executor side of
the ALP SDK model pipeline. tan wraps alp_cli model --format json payloads
in the standard {command, ok, exitCode, project, data, issues} envelope, and
adds the structured ModelSub subcommand group: check · zoo · add · prep ·
run · ab (plus the existing build / list / info / doctor). Single integration
branch so the whole surface lands on dev as one testable PR.

• crates/tan-cli/src/cli.rs — ModelSub variants + per-subcommand arg structs
  (--board, etc.).
• crates/tan-cli/src/commands/model.rs — sub_argv mapping to the
  alp_cli model argv tail; is_wrappable; module-doc header.
• --board forwarded through every board-scoped subcommand.
• CHANGELOG.md [Unreleased] entry.

Scope

Rust CLI (tan-cli) — build-plan/envelope consumer only; no tan-core domain
change. Follows the build-plan CONSUMER contract; keeps the envelope byte-stable.

Test plan (run on this branch)

• cargo fmt --all --check: clean
• cargo clippy --all-targets -- -D warnings: clean
• cargo test: green (see below)
• cargo build --all-targets: builds

Supersedes

Folds the previously-stacked slice PRs (all closed): #47 (envelope), #50 (check),
#51 (zoo), #53 (prep), #54 (measure). Companion branches: alp-sdk +
alp-sdk-vscode integration/model-edge-ai.

[hkngln]

Heads-up: this went to CONFLICTING after #60 and #63 landed on dev just now (dev is at 7660036).

Only CHANGELOG.md actually conflicts — verified with git merge-tree --write-tree origin/dev origin/integration/model-edge-ai:

Auto-merging CHANGELOG.md
CONFLICT (content): Merge conflict in CHANGELOG.md
Auto-merging crates/tan-cli/src/cli.rs      <- clean
Auto-merging crates/tan-cli/src/main.rs     <- clean

cli.rs and main.rs are shared but merge cleanly: #63 added RenodeArgs.core (a different struct from ModelArgs) and #60 added mod venv; to the module list, both away from this PR's hunks. So it is a one-file, ## [Unreleased]-section resolution — keep both your ### Added entry and the two that landed.

What landed, for context:

• fix(flash): resolve west from the workspace venv, like build does #60 — tan flash resolves west from the workspace venv like tan build does; the venv helpers moved out of commands::build::workspace into a new crates/tan-cli/src/venv.rs.
• feat(renode): --core <CORE_ID> to boot one slice of a multicore project #63 — tan renode --core <CORE_ID>, plus dropping --hide-monitor from the Renode argv (Renode 1.16.1 refused it and the smoke silently passed while nothing booted).

Sorry for the churn — merged in that order because the flash fix was blocking local bring-up.

[alpCaner]

Parked until the release chain is out. Please do not rebase this in the interim.

This is one leg of a three-repo trio that has to land atomically:

• alplabai/alp-sdk#933
• alplabai/tan-cli#58
• alplabai/alp-sdk-vscode#366

They cannot land separately: #366 shells tan model subcommands that only exist once #58 lands, so merging the extension leg first ships a command surface calling into a binary that does not have it.

Current state of all three: CONFLICTING. #366 is additionally author-declared not ready, and #58 is four days stale with checks to match.

Why parking beats rebasing right now

The release chain is alp-sdk v0.14.0 -> tan-cli v0.4.1 -> alp-sdk-vscode v0.4.0, and it is in flight — alp-sdk v0.14.0 is gating now.

Rebasing a conflicting three-repo feature into a live release chain is how releases slip. Every rebase re-runs three CI suites, and a trio that must land atomically cannot land at all while two of its legs are red — so the rebases buy nothing and cost review attention that the release needs. tan-cli dev has moved substantially this week (the workspace guard #185, the launch.json splice #184, the contract-code registration #209, the first-blink CI job #215), so a rebase today would be superseded again before the trio is mergeable.

Plan

Pick all three up as a unit, with a single coordinated rebase, once the three tags are cut. At that point dev on each side is stable, the conflicts are resolved once rather than per-leg, and the trio can land atomically the way it has to.

Nothing here is a judgement on the work — it is a sequencing call about when the rebase is worth spending.