render_experiment_docs.py's RVQ deploy-artifact list may reference sample_stimulus.npz when only sample_data.npz exists

[apage224]

Summary

Flagged by Copilot review on PR #59 (unrelated to that PR's actual diff -- pre-existing).

`scripts/render_experiment_docs.py`'s "Deploy Artifacts" section for RVQ experiments lists `sample_stimulus.npz` as one of the canonical local deploy artifacts. But `compressionkit/export/deploy.py::export_for_deployment()` (called directly by the RVQ training recipe, `compressionkit/recipes/base_rvq.py`) only ever writes `sample_data.npz` locally -- `sample_stimulus.npz` is only produced by the separate `compressionkit/experiments/repackage.py::repackage_rvq_golden()` path (used by `compressionkit golden repackage `), via its own `export_stimulus_npz()` call.

So depending on which code path produced a given local `results//deploy/` directory:

• Trained via the normal recipe (`base_rvq.py`) only: has `sample_data.npz`, no `sample_stimulus.npz`.
• Repackaged via `compressionkit golden repackage`: has both.

`sample_stimulus.npz` IS always the correct name for what ends up published to HuggingFace (`RVQ_HF_FILE_RENAMES` maps `sample_data.npz -> sample_stimulus.npz`), so the doc text isn't wrong about the published artifact name -- it's ambiguous about whether it's describing local or published naming, and every currently-regenerated local deploy dir happens to have both files (since all 11 RVQ goldens were repackaged via the `golden repackage` path in PR #59), which is masking the inconsistency.

Proposed follow-up

• Decide: should `render_experiment_docs.py`'s "Deploy Artifacts" section describe local (`results/.../deploy/`) naming or HF-published naming consistently for every family (currently it mixes both: `encoder.tflite`/`decoder.tflite` are local names, but `sample_stimulus.npz` is the published name)?
• Consider whether `export_for_deployment()` itself should always write `sample_stimulus.npz` (or a symlink to `sample_data.npz`) so the "produces the canonical edge deploy package" claim is accurate regardless of which code path (training recipe vs `golden repackage`) built the package.

Context

Found during Copilot review of PR #59 (RVQ encoder/decoder .keras publishing). Not a regression from that PR -- the sample_stimulus.npz bullet point already existed before, and this issue is about a pre-existing local-vs-published naming ambiguity in that same generated doc section.