feat(providers): add Yolo-Auto compatible host

[harryvgiunta]

Summary

Adds Yolo-Auto as a data-driven compatible-host descriptor. Yolo-Auto is a flat-rate OpenAI Chat Completions gateway, so no ProviderKind variant is needed — the wire is ordinary Chat Completions (see #6289 for the wire survey this follows).

The row ships endpoint, credential env (YOLO_AUTO_API_KEY), and a bootstrap model (yolo) only, with discovery: models_endpoint and no compiled roster — the same shape every other compatible host uses. Model ids stay out of the compiled catalog by design.

Billing: the gateway sells flat-rate subscription tiers and publishes no per-token catalog, so billing_surface_for_route classifies the route Unknown (the ApiProvider::Custom path). The route is recorded in the reviewed-omission inventory under unknown_billing_basis with an explicit note that a source review of the billing basis is owed, rather than guessed into subscription or local-no-bill. Compatible-default breadth moves 7 → 8, covered by the existing breadth test. Docs: one row in the known-good hosts table in docs/PROVIDERS.md.

No new layer: this is one descriptor row, one inventory entry, one docs row, one count update. No behavior changes.

Testing

• cargo fmt --all -- --check — clean.
• python scripts/check-provider-registry.py — passed.
• cargo test -p codewhale-config --lib descriptors — passed (2 tests).
• cargo test -p codewhale-tui --lib pricing::default_coverage_tests — passed (1 test).
• cargo test -p codewhale-tui --lib -- picker_lists_all_providers setup — passed (241 tests), covering the picker/setup surfaces that consume bundled_provider_descriptors().
• cargo clippy -p codewhale-config -p codewhale-tui --all-targets --all-features --locked with the CI allow list — the config crate is clean. The tui run reports 29 errors; none are in files this PR touches (all four changed files are JSON/data plus one --lib test count), and they are all newer-lint findings on pre-existing code (collapsible_match, nonminimal_bool, needless_return under clippy 1.95). Not introduced here; CI pins the toolchain and is authoritative.

Platform caveats, stated plainly:

• The Windows GNU toolchain on this machine cannot link the codewhale-tui test binary (ld: cannot find -lshlwapi, missing mingw import lib; CI is MSVC). The tui test runs above used an empty stub archive on -L to get past the link step only; compilation and all 242 assertions are real.
• The two unfulfilled_lint_expectations errors under [workspace.lints.rust] warnings = "deny" reproduce identically on pristine main with rustc 1.95.0 — they pre-date this change. The local test runs temporarily used warnings = "warn"; Cargo.toml is untouched in the diff.

Checklist

• This PR adds a new layer/module/abstraction — it names or deletes the layer it replaces
• Updated docs or comments as needed
• Added or updated tests where relevant — the compatible-default breadth assertion and the reviewed-omission inventory are the test surface for a descriptor row; the descriptor itself is data validated by check-provider-registry.py and the config crate's descriptor tests
• Verified TUI behavior manually if UI changes — N/A, no UI change (descriptor surfaces through the existing picker once CI links the binary)
• Harvested/co-authored credit uses a GitHub numeric noreply address

[github-actions]

Thanks @harryvgiunta for taking the time to contribute.

This repository is observing a maintainer-managed PR intake gate in dry-run mode, so this pull request is staying open. This note helps maintainers prepare the allowlist before any enforcement is considered.

Please read CONTRIBUTING.md for the expected contribution shape. A maintainer can grant recurring PR access by commenting /lgtm on a pull request.

[harryvgiunta]

Gate acknowledged. Rebased onto current main (6c8f242) and reshaped the description to the PR template. Per the pre-push section I also ran the strict release-form clippy (--all-targets, CI allow list): config crate clean; the tui findings are all in files this PR does not touch and are listed honestly in the Testing section rather than reported as green. Single-purpose change: one descriptor row, one inventory entry, one docs row, one count assertion. Happy to hold until the allowlist decision; no action needed from maintainers beyond /lgtm if it is granted.