Harden generated media tags and batch tag merge

[brianmeyer]

Summary

• strip fenced markdown wrappers before parsing generated media tags
• merge duplicate search_batch tag sets deterministically across contributing hits
• add regressions for fenced JSON tag output and duplicate batch-hit tag merging

Testing

• pytest -q tests/test_storage.py -k "generated_media_tags_strip_fenced_json or memory_lookup_surfaces_media_tags or index_image_stores_caption_in_text_body or index_video_stores_video_caption_in_text_body"
• pytest -q tests/test_search_batch.py -k "same_document"
• pytest -q tests/test_storage.py tests/test_search_batch.py tests/test_search_pipeline.py

Closes REC-192
Closes REC-193

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: bf4ec6fa1d

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Preserve query order when merging duplicate result tags

The new tag merge path appends duplicate hits in whatever order all_results is iterated, but all_results is populated via as_completed, so completion timing can reorder queries between runs. In batches where multiple queries hit the same filepath with different tags, tags=_merge_tags(data['results']) will produce different tag orders for identical inputs, which breaks reproducibility and can make deterministic-order assertions flaky.

Useful? React with 👍 / 👎.