QVAC-22630 chore: release @qvac/model-fit 0.6.0

[simon-iribarren]

🎯 What problem does this PR solve?

@qvac/model-fit 0.5.0 is the published version, and #3930 has since merged to main — the disposable-process protocol v2, the completion/embedding load normalization behind it, and the review fixes that came out of it. None of it is installable yet.

📝 How does it solve it?

Bumps packages/model-fit/package.json to 0.6.0 and records the release in packages/model-fit/CHANGELOG.md. No source change.

A minor bump, not a patch or a breaking one:

• It adds public surface — protocol v2 with an explicit loadKind, encodeFitLlamaProcessRequest, and the FitLlamaResult / FitLlamaReason types carrying the unsupported-config outcome.
• Nothing published is removed or narrowed. FitResult gained unsupported-config and then lost it again inside QVAC-22630 feat[api]: add config normalization to model-fit addon #3930, both before release: index.d.ts on release-model-fit-0.5.0 already declares the status-2 reasons as 'model-unreadable' | 'no-backend-device', exactly what 0.6.0 ships. fitParams()'s contract is byte-identical to 0.5.0's.
• Protocol v1 is unchanged and still accepted.

🧪 How was it tested?

Verified at the merge commit before cutting the branch:

• ctest --test-dir build --output-on-failure — 3/3
• npm run test:process — process.test.js 21/21, llama-config.test.js 17/17 (328 asserts)
• bare test/integration/fit.test.js — 27/27 against a real GGUF
• npm run test:node — 2/2
• npm run test:types — typecheck, declaration consumer and check:generated
• npm run lint, clang-format, clang-tidy (0 user-code warnings)
• CI on QVAC-22630 feat[api]: add config normalization to model-fit addon #3930 at its final head: 44 checks pass, including the full seven-runner desktop integration matrix under run-desktop-addon-tests.

This PR itself contains no source change, so it inherits that state.

🔌 API Changes

Additive. fitParams() and FitResult are untouched relative to 0.5.0.

import {
  encodeFitLlamaProcessRequest,
  parseFitProcessResponse,
  resolveFitProcessRunnerPath
} from '@qvac/model-fit/process'
import type { FitLlamaResult } from '@qvac/model-fit/process'

const request = encodeFitLlamaProcessRequest('completion', {
  modelPath: '/absolute/path/model.gguf',
  params: { device: 'gpu', 'ctx-size': '4096', 'gpu-layers': '40' },
  marginMiB: 1024
})

// Send `request` as the runner's single stdin line, then:
const response = parseFitProcessResponse(JSON.parse(line))
if (response.status === 'completed') {
  const result: FitLlamaResult = response.result
  // `unsupported-config` is advisory — it must never be used to deny a load.
}

The raw load-config fitter is deliberately not in-process public API: ./binding.js re-exports paramsFit only, and the runner reaches the fitter through the unexported ./binding-internal.js.

📦 Release

• Base branch release-model-fit-0.6.0, cut from main at 975b36ea3.
• Merging this triggers the GPR publish; the backmerge into main follows and triggers npm.

[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):

• ./.github/actions/release-merge-guard
• ./docs/website
• ./packages/ggml-coload-smoke
• ./packages/fabric/test/integration
• ./packages/inference-addon-cpp/mobile
• ./packages/sdk/e2e
• ./packages/llm-llamacpp/benchmarks/performance
• ./packages/llm-llamacpp/benchmarks/server
• ./packages/vla-ggml/sim/server
• ./packages/embed-llamacpp/benchmarks/performance
• ./packages/embed-llamacpp/benchmarks/server
• ./packages/asr-ggml/benchmarks/server