Feature/SUPeR expansion (expand other 13 DBs from NIE pseudobulk to other DBs for MVP)

[VinLau]

Registers 13 more SUPeR Viewer dataset pairs alongside arabidopsis_NIE_*, which shipped in #330. Pseudobulk databases are catalog entries only; UMAP databases move from a hand-written model file to a registry dict (umap_dynamic.py) that essentially is the same Model simply 'renamed'. First @asherpasha you'll need to upload the full DB set and then we can merge this PR in (I'll do some testing as well).

1. Summary

Four commits, each reviewable on its own.

Commit	What it does
d6a605c	Replaces api/models/arabidopsis_NIE_umap.py and the if/elif in UMAPUtils.get_tables with api/models/umap_dynamic.py: a UMAP_DATABASES dict mapping database → species, and a factory that builds the coords and expression models per database. Modelled on efp_dynamic.py. Only arabidopsis_NIE_umap is registered here.
6f871f1	Adds 26 fixture dumps and their CI wiring — one SQLALCHEMY_BINDS entry in config/BAR_API.cfg and one seed line in config/init.sh per database.
496ebb8	Adds 13 pseudobulk catalog entries to combined_master.json, each copying arabidopsis_NIE_pseudobulk field for field except species/gene_id_pattern (different only for rice_OW_pseudobulk). pseudobulk_std.assigned_databases 1 → 14. Updates the expected set in test_gene_expression.py to the explicit 14 names.
bafa6fe	Registers the 13 new UMAP databases and adds tests/resources/test_superviewer_databases.py.

d6a605c is behaviour-preserving. tests/resources/test_umap_gene_expression.py is byte-identical to dev and still passes, so the refactor changed no route, response or error.

UMAP databases get no catalog entry: they have no sample_data table, so an entry would build a model for a table that doesn't exist. The registry dict is a code-level list instead.

Test literals in test_superviewer_databases.py were taken from the committed fixtures and are checked against fixture_manifest.json; the tests never read the manifest at runtime. Suite from an empty MySQL seeded only through init.sh: 122 passed, 91 subtests passed, 3 failed. The 3 failures are dev's existing baseline (mfinder exit 127, two Redis refusals on localhost) and this PR touches none of them. flake8 clean.

Both endpoints were also exercised directly against a local instance (47 requests): row counts and Mean_CTRL values match the fixtures for all 14 pseudobulk databases, and across all 13 UMAP databases every expression key is a coordinates key with no gap cell ever returned. That is fixture data, so it does not substitute for the staging check in the merge gate.

2. Merge checks

Do not merge on staging until both pass. Both need the 26 databases on prod first; I can run them.

1. Schema check. Run Vin's local verify_master_against_prod.py on staging server MySQL against live information_schema. The 13 new entries copy schema_verified: true and schema_source: "prod_information_schema" from the NIE entry, but so far those claims rest on the dump files, not on prod. The script covers only the 13 pseudobulk entries — the UMAP databases aren't catalogued, so nothing verifies them this way.
2. Staging test. For each new pseudobulk database, the fixture genes' Mean_CTRL value and value_std must equal the literals in tests/resources/test_superviewer_databases.py. For each new UMAP database, the coordinates response and each fixture gene's response must return 200.

3. Databases (reference)

26 new databases. Names are case-sensitive; note the capitalised OW.

arabidopsis_flower_lee_pseudobulk       arabidopsis_flower_lee_umap
arabidopsis_root_shahan_pseudobulk      arabidopsis_root_shahan_umap
arabidopsis_rosette_21d_lee_pseudobulk  arabidopsis_rosette_21d_lee_umap
arabidopsis_rosette_30d_lee_pseudobulk  arabidopsis_rosette_30d_lee_umap
arabidopsis_seed_0d_lee_pseudobulk      arabidopsis_seed_0d_lee_umap
arabidopsis_seed_martin_pseudobulk      arabidopsis_seed_martin_umap
arabidopsis_seedling_12d_lee_pseudobulk arabidopsis_seedling_12d_lee_umap
arabidopsis_seedling_3d_lee_pseudobulk  arabidopsis_seedling_3d_lee_umap
arabidopsis_seedling_6d_lee_pseudobulk  arabidopsis_seedling_6d_lee_umap
arabidopsis_shoot_zhang_pseudobulk      arabidopsis_shoot_zhang_umap
arabidopsis_silique_lee_pseudobulk      arabidopsis_silique_lee_umap
arabidopsis_stem_lee_pseudobulk         arabidopsis_stem_lee_umap
rice_OW_pseudobulk                      rice_OW_umap

4. Known data gaps/inconsistences, not addressed here

Counts below come from a full-dump scan (21 Sept 2026), not from anything in this PR. The committed fixtures use pattern-valid gene keys only, so none of this affects the tests.

• shoot_zhang — 9,570 symbol-keyed genes, some containing spaces or slashes. Not reachable through the API's gene ID pattern. Separately, the UMAP route declares <string:gene_id>, which does not match slashes, so a slash-bearing key is rejected by Flask routing (404) before any handler runs. The pseudobulk route uses <path:gene_id> and has no such limit.
• root_shahan — 21,437 lncRNA keys (AthLNC + six digits), plus 7 ATMG…-N duplicates. Real loci with no AGI, so not reachable through the Arabidopsis pattern.
• NIE — 6,423 symbol-keyed genes; 11 AGI keys carrying 138 rows (two genes each); one nan row in the UMAP table.
• seed_martin — some data_bot_id values end in an apostrophe, e.g. 3DAP ii1' (integument-prime names). The committed fixture has 3 such values across 9 rows. They're backslash-escaped in the dumps, so anything parsing a dump or rendering sample names must handle them — a naive single-quote parser silently drops those rows.

Reaching the symbol- and lncRNA-keyed genes needs a decision about gene ID patterns that would affect 30+ databases, so it's deliberately left out of this PR. Note that widening a pattern alone would not be sufficient for slash-bearing keys on the UMAP endpoint — its route converter would have to change too.

5. Contributor note

docker-compose.yml bind-mounts only api/, tests/ and config/databases/. combined_master.json, config/BAR_API.cfg and config/init.sh are baked into the image, and app.py runs without a reloader. After changing any of those three, rebuild:

docker compose up -d --build

Otherwise local tests run against stale copies — the usual symptom is a newly catalogued database returning "Invalid species or gene ID".