feat(engine): consolidate the engine PRs, with on-device-labelling docs and compat tests

[flxkm522]

What this is

One consolidated PR for the redrob-code engine work, reviewed and refactored, plus the on-device-labelling documentation and compatibility tests (formerly #81, now folded in). The Electron/desktop app and the build-off / Quality-Lab stack are the separate alpha/desktop drafts and are not part of this.

Part A — Engine consolidation (was seven PRs, now closed)

Was	What
#67	insights route + /v1/chat/completions (serve the OpenAI face for real; forward x-redrob-session)
#68	regenerate the JS SDK that had fallen behind the routes
#69 → #70 → #71 → #73 → #74	image_generate (edit / reference images), speech_generate (mp3), POST /redrob/speech, POST /redrob/transcribe
#79	shared on-device labellers (route-labeller, work-labeller) as source workspace packages

Integration fix: the chat-completion handler reads LLMClient inside the request, so it was a per-request requirement of the embedded API that Layer.provide(serviceLayer) did not discharge — it leaked and broke the untouched sdk-next caller. LLMClient is global, so it is now provided once via HttpRouter.provideRequest inside makeRoutes (packages/server/src/routes.ts).

Lean/doc pass (behaviour-preserving): constant-derived size messages; idiomatic [...text].length; a tightened describe() union guard; a corrected plugin/host comment; /redrob/speech+/redrob/transcribe documented in LOCAL-ENGINE-API.md; and route-coverage scenarios added for the two /redrob/* routes.

Part B — On-device labelling: docs + compatibility (answering the CTO review)

• docs/ON-DEVICE-LABELLING.md (new) — the single reference for why there is encoder code (we wrap pinned OSS models — distiluse, multilingual-e5-base — not our own encoders, for on-device privacy + one implementation across products), where the tensors live (only the small trained layers are ours and in git: work-head.json 159KB, prototypes.json 232KB; base weights are hash-pinned downloads), the model-free-CI vs opt-in-eval story, and the backwards-compat model. Carries the pending checklist to move the e5 release asset from redrob-cowork to redrob-code (SHA pins unchanged; the URL is not flipped here because a not-yet-uploaded release would 404 every download).
• httpapi-chat-completion.test.ts — two new v1-contract cases: the newer optional fields (max_completion_tokens, reasoning_effort, user) stay accepted, and unknown request fields are ignored, not rejected (a Console-style redrob block + a bogus key). The second pins the exact forward/third-party-compat guarantee and fails CI if the schema ever rejects excess properties.
• test.yml — the labellers (linux) comment now spells out that the unit suites are model-free and CI-safe, and the real encoders are needed only by the opt-in eval/model dev commands (off the PR path).

Verified

bun turbo typecheck 20/20; bun install --frozen-lockfile clean; engine tool/server tests + the 7 chat-completion contract cases pass; core 6/6; work-labeller 77/77.

(This supersedes #81, whose three files are now the single commit here.)

[its-janghoon]

Closed by the maintainer: engine PRs are not being taken right now. Branch kept.