release: publish and record corrected RVQ v1.1 bundles

[apage224]

RVQ deployment packages previously contained EMA accumulators as extra codebooks, and the HF publisher dropped independent references and checksum files. Eleven corrected v1.1 bundles are now published and verified by immutable revision. This PR records their provenance, preserves the complete canonical deploy contract when staging, and directs current single-stage examples and the golden registry to the corrected track.

Refs #66. This completes the corrected-export publication portion; the remaining product-readiness checklist stays open.

Changes

• Preserve canonical deploy files plus legacy HF aliases; regenerate checksums after README and alias changes.
• Render corrected-export provenance and explicitly label historical physiological scorecards.
• Promote the 11 single-stage RVQ entries to v1.1. Keep unqualified entropy priors on v1.0 and reject publication across a parent release-track mismatch.
• Record all 11 pinned repositories, checksum-manifest hashes, packet geometry, source checkpoint hashes and fresh precision reports in releases/rvq-v1.1.json.
• Update runtime examples and release documentation. Downstream selection/pinning is handled by the compressionkit-dev #2.

Validation

• All 11 local, staged and revision-pinned downloaded packages pass strict validation on 10 independent training-reference frames each; no validation warnings.
• The complete registered-package sweep (golden validate-all --strict-release) passed 33/33 unique packages: 11 corrected RVQ candidates and 22 existing SPIHT/hybrid packages.
• All 11 corrected codebook headers compile as C99.
• 440 complete-packet round trips across FP32, INT8, FP16 and INT16x8; downstream vendored runtime independently passed another 44 round trips across all 11 published revisions.
• Every downloaded payload matches the published checksum manifest. Original local artifacts and v1.0 repository heads remain unchanged.
• Full devcontainer suite: 445 passed, 78 optional data/artifact skips. Ruff and formatting pass.

ECG 64x INT8 has 13.64% P90 parity PRD and PPG 32x INT8 has 11.31%, above the 10% recommendation; model cards identify the higher error and FP32 remains the default. No retraining, new physiological evaluation, MCU parity measurement or entropy-prior qualification is claimed. Model weights and generated deployment artifacts remain git-ignored.

[apage224]

Final validation at bc60ced: GitHub CI and all Astro documentation checks pass, including migration, links and rendered layout. The registered-package strict sweep passed 33/33 packages; the full source suite passed 445 tests (78 optional artifacts/data skips). All 11 corrected codebook headers compile as C99.

The 11 v1.1 bundles are already published and checksum-verified by exact revision. Local/staged/download strict validation, 440 release packet round trips and 44 downstream packet round trips passed. Historical v1.0 repositories and local originals were preserved. Companion downstream reference update: https://github.com/AmbiqAI/compressionkit-dev/pull/2.

This PR is ready for review and merge. No model weights or generated deployment payloads are tracked in Git; only source, documentation and release inventory metadata were committed.