Add collection facet to explorer (e.g. OpenContext PKAP) (#243)

[rdhyee]

Resolves #243.

Adds a first-class collection dimension to the explorer: filter to a named
SamplingSite label (e.g. the OpenContext project "PKAP Survey Area") and
layer the existing material / context / object_type facets on top.

Why this design (additive)

"Collection" identity lives on SamplingSite entities, reached only by the
MaterialSampleRecord → produced_by → SamplingEvent → sampling_site → SamplingSite traversal — never on the sample rows the explorer renders. Doing
that array-join live in DuckDB-WASM is the documented in-browser bottleneck, so
membership is precomputed. The current sample_facets_v2 / facet_summaries / facet_cross_filter build pipeline isn't in any repo, so rather than risk
regenerating those, this feature is strictly additive — two new files that
touch nothing existing:

• collections.parquet — dimension (collection_id, label, source, n_samples, centroid_lat/lng, bbox). 61,695 rows, ~3 MB. Powers the top-N
  checkboxes, the search box, and the Featured-Collections preset cameras.
• sample_collections.parquet — membership (pid → collection_id). ~13 MB.
  The filter appends a second pid IN (SELECT … ) subquery in
  facetFilterSQL(), exactly parallel to the existing facet predicate.

A "collection" = a SamplingSite label (≈1,336 site rows share "PKAP Survey
Area"), keyed by a stable hash of (source, label). Verified: PKAP = 15,446
samples.

What's in the PR

• scripts/build_collections.py — builds both files from /current/wide.parquet.
• explorer.qmd — dual-UX collection facet (top-N checkboxes + search-the-tail
  for the ~60K long tail), ?collection= URL param wired through the existing
  facet lifecycle (applyQueryToFacetFilters / writeQueryState /
  handleFacetFilterChange) and the facetFilterSQL() chokepoint.
• collections.qmd — Featured Collections page upgraded to identity-based
  &collection=<id> links + camera fly.
• EXPLORER_STATE.md, data.qmd — document the new param and files.
• tests/test_collections.py — Collections page + explorer facet-DOM checks.

⚠️ Merge gate — requires R2 upload first

The facet is inert until the two files are live on data.isamples.org:

• Run python scripts/build_collections.py --out-dir <dir> --snapshot 202604
• Upload isamples_202604_collections.parquet + isamples_202604_sample_collections.parquet to R2 (behind the data.isamples.org Worker)
• Verify live: open explorer.html?collection=dd74c71982da0e21 → PKAP samples; layer a material facet to confirm it narrows
• Run tests/test_collections.py against the deployed site

Known limitations (v1)

• Collection facet counts are the collection's static total — not
  cross-filtered against other facets (no cross_filter cache for collections).
  The dots and table do respect the filter. Documented in EXPLORER_STATE.md.
• Like the other facets, collection filtering applies at neighborhood/point zoom,
  not to zoomed-out H3 clusters (same #facetNote caveat).

🤖 Generated with Claude Code

[rdhyee]

Rebased onto upstream/main (was 41 commits behind) to resolve conflicts with the #305 facet-count stack. All conflicts were in explorer.qmd, in the four spots where this branch's collection-facet wiring sat next to the #305 facet-tree/mask/index machinery — composed both sides rather than picking one:

• Data-URL block: kept Meta: facet-count correctness & coherence (counts beyond the single-filter cube) #305's vocab_labels_202608/facet_tree/masks/index URLs and appended this branch's collections_url/sample_collections_url/COLLECTION_FACET_TOPN.
• applyQueryToFacetFilters(): kept Meta: facet-count correctness & coherence (counts beyond the single-filter cube) #305's tree-visual sync loop plus this branch's setCheckedValues('collectionFilterBody', ...) restore.
• facetFilterSQL() (the chokepoint): kept Meta: facet-count correctness & coherence (counts beyond the single-filter cube) #305's parts-array structure (bitmask/membership + flat facets) and added the collection predicate as one more parts.push(...) entry, replacing this branch's older string-concatenation version of the same function.
• Facet rendering: kept Meta: facet-count correctness & coherence (counts beyond the single-filter cube) #305's tree-vs-flat renderTreeFacet/renderFilter loop and appended this branch's collection top-N + search-the-tail block right after it, before applyFacetCounts.
• heatmapFilterHash(): kept both the explorer: architectural direction — make filter semantics coherent across all surfaces #234 search-token field and this branch's collection field.

No EXPLORER_STATE.md/_quarto.yml/data.qmd conflicts — those merged cleanly.

Tests: npm run test:unit — 13/13 pass. tests/test_collections.py needs a live rendered site (not the prod fallback, since collections.qmd/explorer.qmd changes aren't deployed there yet) — ran it against a local quarto preview server (ISAMPLES_BASE_URL=http://localhost:5860) and all 7 tests pass.

Still draft — leaving as-is since the R2 upload of isamples_202604_collections.parquet/isamples_202604_sample_collections.parquet hasn't happened yet per the PR description's merge gate.