QVAC-21691 feat[api]: SDK-side support for the Python full-fidelity client

[lauripiisang]

🎯 What problem does this PR solve?

The Python SDK client (separate PR, stacked on this one) needs behaviour that
only the @qvac/sdk worker can provide once, so every language binding shares
it instead of re-implementing (and drifting). This PR is the @qvac/sdk side
of that work, split out so it reviews on its own.

🛠️ How does it solve it?

• Worker-orchestrated completion — a new completionOrchestrate duplex
  method runs the multi-turn tool loop + MCP in the worker and emits
  toolCallback frames; the client only routes each callback to a
  user-supplied handler. The existing completion() API is unchanged.
• Translate detection → worker — server/bare/ops/translate.ts
  auto-detects the source language when from is omitted (LLM), so the JS and
  Python clients share one detector instead of each shipping its own (they
  disagreed on the same input). The LLM server-param refine now requires only
  to; the wire from was already optional.
• Plugin contract export — scripts/export-plugin-contract.ts emits a
  per-plugin contract so custom plugins get typed clients.
• Data exporters for the generated Python client —
  contract/model-type-maps.json (alias/engine → canonical model type) and
  contract/error-codes.json (error-code registries), so the Python client
  generates those tables instead of hand-copying them; contract:check fails
  if they drift.
• Cross-client conformance corpus — e2e/conformance/cases.json with a
  direct-SDK JS runner (run.mjs); the Python client runs the same cases, so
  the two can't diverge on covered behaviour.

🔌 API changes ([api])

New public duplex method completionOrchestrate (additive; nothing removed):

const run = completion({ modelId, history, tools, orchestrate: true })
for await (const event of run.events) { /* ... */ }
const final = await run.final

✅ Testing

• bun run build (lint + typecheck + compile), contract:check, and
  bun run test:unit all pass.
• completionOrchestrate handler + plugin-contract unit tests added.
• run:local:desktop --filter translation-llm — 9/9 against the rebuilt worker
  (covers the from-omitted autodetect path).

📝 Notes

The Python client PR (packages/sdk-python) is stacked on this branch — merge
this first, then that one retargets to main.

[github-actions]

Review Status

Current Status: ❌ PENDING
Approvals so far: none

Pending reviews: Needs 1 Management or Team Lead, and 1 more from Management, Team Lead, or Member.

[github-actions]

License compliance — clean

No new dependency license findings in this PR.

Warn-only (shadow) mode — this check does not block merges yet.

Updated automatically by the canonical license compliance workflow.

NOTICE presence (advisory)

Missing NOTICE (advisory, does not block):

• ./packages/embed-llamacpp/benchmarks/server
• ./packages/embed-llamacpp/benchmarks/performance
• ./packages/ggml-coload-smoke
• ./packages/llm-llamacpp/benchmarks/server
• ./packages/llm-llamacpp/benchmarks/performance
• ./packages/fabric/test/integration
• ./packages/transcription-whispercpp/benchmarks/server
• ./packages/transcription-parakeet/benchmarks/server
• ./packages/sdk/e2e
• ./packages/vla-ggml/sim/server
• ./.github/actions/release-merge-guard
• ./docs/website

[github-actions]

QVAC E2E — android — ✅ all tests passed (91/103, 1773s)

Config: suite=smoke · filter=(none) · exclude=(none)
Device pool: Google Pixel 10 Pro - Android 16 — ☀️ highly available
View run · Artifacts: reports · Device Farm logs

[github-actions]

QVAC E2E — ios — ✅ all tests passed (91/103, 796s)

Config: suite=smoke · filter=(none) · exclude=(none)
Device pool: iPhone 16 Pro - iOS 18 — ☀️ highly available
View run · Artifacts: reports · Device Farm logs

[github-actions]

QVAC E2E — windows — ✅ all tests passed (103/103, 415s)

Config: suite=smoke · filter=(none) · exclude=(none)
View run · Artifacts: reports

[github-actions]

QVAC E2E — linux — ✅ all tests passed (103/103, 277s)

Config: suite=smoke · filter=(none) · exclude=(none)
View run · Artifacts: reports

[github-actions]

QVAC E2E — macos — ✅ all tests passed (103/103, 220s)

Config: suite=smoke · filter=(none) · exclude=(none)
View run · Artifacts: reports

[lauripiisang]

Superseded by #3354 — the SDK contract, generated Python client, and worker behaviour are CI-coupled and can't be split into a clean SDK-only PR, so the full-fidelity Python work ships as one PR (#3354).