QVAC-23075 feat[api]: bump qvac-fabric to 10069.1.0 across consumers

[iancris]

🎯 Problem

qvac-fabric 10069.1.0 adds VisionPsy Nano support and its Flash preprocessing rule
(tetherto/qvac-fabric-llm.cpp#205).
All 7 fabric consumers are still pinned to a version>= floor of 10069.0.0, so none of them can
resolve the new fabric.

📝 How

One bundled PR carrying the whole consumer-side rollout for all 7 fabric consumers:

• vcpkg.json — qvac-fabric version>= floor 10069.0.0 → 10069.1.0
• package.json — version bump (all 7 are pre-1.0, so all take the minor path)
• CHANGELOG.md — dated ## [<version>] entry recording the fabric bump

Package	Version
@qvac/embed-llamacpp	0.32.0 → 0.33.0
@qvac/fabric	0.4.0 → 0.5.0
@qvac/llm-llamacpp	0.43.0 → 0.44.0
@qvac/model-fit	0.1.0 → 0.2.0
@qvac/ocr-ggml	0.16.0 → 0.17.0
@qvac/translation-nmtcpp	0.8.0 → 0.9.0
@qvac/vla-ggml	0.19.0 → 0.20.0

No C++ or JS source changes — the fabric bump is API-neutral for every consumer. No
default-registry.baseline changes.

Depends on the registry publish: tetherto/qvac-registry-vcpkg#317

CI will fail until #317 merges, because version>=: 10069.1.0 cannot resolve before then. Expected
— CI needs a retrigger once the registry PR lands.

🧪 Tested

• Fabric tag v10069.1.0 created at 50713f4dfc54b1fe09826a343527e83d90f5df62, the head of the
  temp-10069 release branch (the merge commit of fix(ci/ocr): fix vcpkg cache not saving by splitting restore/save #205).
• Release archive SHA512 computed and cross-checked with both vcpkg hash and sha512sum.
• The registry port built from the new tag against a clean, isolated vcpkg install root
  (x64-linux, [core,gpu-backends,llama]): archive downloaded, SHA512 accepted, build and install
  succeeded.
• Phase A overlay-validated all 7 consumers against the fabric branch before the tag was cut; that
  overlay has since been reverted from PR QVAC-23075 feat[api]: accept image_no_upscale in the addon load config #3725.
• Branch cut fresh from origin/main; verified no vcpkg-overlays/ports/qvac-fabric subtree, no
  overlay-ports key anywhere in the repo, and no vcpkg-configuration.json drift.

⚠️ Breaking

None. No addon API changes in any of the 7 packages — each changelog entry records the dependency
bump only.

[iancris]

Superseded by #3725 — closing.

This PR should not have been opened. The rollout was run with --on-top-of-pr #3725, which exists precisely so that the consumer code needing the new fabric and the fabric bump land on the same PR. Splitting the bump out into a separate PR recreates the problem that flag was added to solve: #3725's CI keeps building against the old published fabric, and the two only meet after the whole rollout lands.

The identical bump now rides #3725 as commit d524054 — same 7 consumers, same version>= 10069.0.0 → 10069.1.0, same package versions and changelog entries. Registry publish is still tetherto/qvac-registry-vcpkg#317.

One difference worth noting: on #3725 the bump was applied on top of a main merge, because that branch predated the @qvac/llm-llamacpp 0.43.0 release. The version in this PR was derived from origin/main (0.43.0 → 0.44.0), which was correct here but would have conflicted with #3725's 0.42.0 base — so the commit was re-derived there rather than cherry-picked.

The /rollout-phase-b skill has been corrected so --on-top-of-pr no longer opens a separate PR.