Send bare Jev model ids in OpenRouter's own form; release 0.7.4

[DevMortimer]

What and why

On the OpenRouter backend, a caller who wrote the TypeSafe default model: "jev-latest" got HTTP 400. OpenRouter names that model ~typesafe/jev-latest and a pinned version typesafe/jev-<major>.<minor>, so the bare TypeSafe id did not resolve.

Change

• backendModelId(backend, model) in src/backends.ts maps a bare Jev id for OpenRouter: jev-latest becomes ~typesafe/jev-latest, and jev-1.13 or jev-1.13.0 becomes typesafe/jev-1.13. Ids that already carry an author (vendor/model), unknown bare ids, and every model on the TypeSafe backend pass unchanged. Mapping is idempotent.
• The client applies the mapping to its default model and to a per-request model in evaluate(). The 1–100 character check still applies to the caller's own id, before mapping.
• /typesafe status derives its Model: line from the same rule instead of a hardcoded jev-latest. The extension uses the TypeSafe backend, so the visible text is unchanged.
• docs/api.md documents the mapping. Not exported from src/index.ts, so the public API is unchanged.
• Release 0.7.4: package.json, package-lock.json, and CHANGELOG.md.

Verification

• npm ci and npm run check pass: build, typecheck, 104 offline tests.
• Offline tests capture the request body through the injected fetch: the default model, jev-latest, jev-1.13, jev-1.13.0, an id with an author, the TypeSafe backend unchanged, and a per-request model on OpenRouter. One test checks the status line.
• Documentation and changelog updated.

Live check: OpenRouter's public model endpoint resolves both ~typesafe/jev-latest and typesafe/jev-1.13, and returns 404 for a version that does not exist. No key was used, and no judgment request was sent.

Compatibility and release notes

• No public export added or removed. TypeSafe backend behavior is unchanged.
• Limitation: the mapping is keyed on the backend name, so a custom backend config passes model ids unchanged.