feat(engy): add engy sync module - #8715

Summary

Adds a sync module for engy, the provider added in #5910. That PR was split so the provider could land first, and this is the module that was taken out of it. Changes since then came from comparing it with the sync modules merged before it. It now keeps every authored field, as the Mistral module does; before, a fixed field list would have stripped v2's new type field. It also validates prices and leaves a stale input/output split to a human (both below).

engy publishes prices, the context window and modalities on a public GET https://api.engy.ai/v1/models (no auth). The module keeps those current. While #5910 was open it caught two real changes: kimi-k3's 30% price rise on 09-03 and qwen3.8-27b's larger window on 09-13.

What it writes and what it leaves alone

• Writes the cost (input, output, cache_read), limit.context and modalities.
• Leaves everything else as authored: the input/output split, reasoning_options, interleaved, any other cost keys, provider, experimental and status. The public list gives only the context window, which on engy is max_input + max_output, so the module never derives the split from it.
• base_model: an authored base_model wins over the slug resolver.
• Modalities: an absent, empty or unrecognised list keeps what the file says, and file maps to pdf. A context_length of 0 or less counts as unreported.
• Stale split: engy's context is max_input + max_output. When context_length no longer equals the authored limit.input + limit.output, the module leaves the file alone and opens a [missing-model] issue (via MissingReasoningOptionsError, as Mistral does for API data that contradicts the file), because the public list can't say which half moved. kimi-k3's window change on 10-02 was exactly this case.
• Bad rows: a price that isn't a non-negative number fails the run rather than writing a free model. A row with an empty id is ignored; engy serves one as of 10-09.
• skipCreates: a new id opens a [missing-model] issue instead of a file, because a created file would ship limit.output = 0. Today the list and the catalogue match, so the first run opens nothing.
• deleteMissing: false: {"data":[]} passes the schema, so retired ids are reported, not deleted.
• Prices are rounded to six decimals, so the hourly run is a no-op when nothing has changed.

The module is registered in sync/index.ts (alphabetical, direct group), with an engy:sync script and an engy Notes section in sync.md. It is not on REVIEWED_REASONING_PROVIDERS.

Test plan

• packages/core/test/engy.test.ts: 25 pass, including the fetch path with an injected fetcher (success and 503) and the registration. Reverting each guard one at a time (split check, price validation, empty-id filter, modality fallback, file→pdf, injected fetch, integer context, field list) fails at least one test. The suite reads the shipped engy files and asserts that a sync round trip leaves them unchanged.
• bun models:sync engy --dry-run, then a real run, then --dry-run again, 2026-10-09 and 2026-10-11: 8 unchanged each time, no files touched; the empty-id row engy served on 10-09 is ignored
• bun validate
• bun run test in packages/sdk: 25 pass
• tsc --noEmit in packages/core: the same 78 errors as on dev, none of them in engy code
• Based on dev 6633e87