Add a validated BGE/PaCMAP taxon overview with full-corpus vector coverage

[realmarcin]

TaxonMech gains the fleet semantic text map using pinned BGE-large-en-v1.5 vectors and actual PaCMAP. The complete 625,960-record source snapshot has validated cached vectors. The published overview displays a deterministic 50,000 taxa and explicitly reports the other 575,960 as omitted from the view. Points open the existing taxon viewer; the complete catalog, AllTheBacteria, StrainInfo and legacy redirects remain available.

The streaming adapter describes taxon names, definitions, rank/domain, lineage and source-qualified synonyms. Each bundle records input hashes, the exact encoder/runtime profile, projection parameters and coverage. Rendering exports current full inputs, validates checksums and the approved generation, then stages the map. Changes to membership or selected text require an incremental cache refresh and new PaCMAP projection before publication. Ordinary CI and rendering validate artifacts without running neural inference or fitting coordinates.

The local-link audit also accepts a directory URL only when it has a real index.html. Its focused cases cover query strings, fragments, missing paths and absent index files while retaining the full taxon-membership and shard-coverage checks. Synthetic specialty-renderer tests now use a temporary repository root, explicit disabled common-map config and a bounded redirect inventory; production full-input validation remains strict.

Validation completed for source 6af4d5137026ef5d443590ebeb49b1ff97e5d8cc:

• Full 625,960-vector cache and source-input checks; actual 128-record enabled-renderer canary.
• Full maintained site render; all 50,000 point routes verified against 2,713 real taxon-detail shards. All 73,774 existing site files preserved apart from the common navigation addition.
• Publication-size gate: 652,600,359 bytes. Actual browser search opened a gzip-backed taxon record and its source-YAML link with no page errors.
• Independent final artifact/documentation review passed. Maintained lint, documentation and raw-provenance checks passed.
• Accept real directory-index routes in the published-site link audit #60 directory-link audit and Isolate renderer fixtures from full-corpus semantic-map preflights #61 bounded renderer fixtures independently reviewed. All original 1,051 selected tests plus 11 new cases are covered across the preserved and resumed runs: 1,059 passes, 3 skips, no unresolved failures.
• Maintained history, strict schema, corpus reproduction, site reproduction, publication-budget and corpus-report gates completed.
• Validation-only CLAW #440 runtime transition and focused current-bundle/staging review completed, preserving the original numerical artifacts and build receipts. All 25 focused adapter/publication tests passed.

The maintained vendored-artifact check passed for all 14 artifacts applicable to TaxonMech at the released pin.

Both required GitHub checks (qc and vendored-sync) passed on final head ec9c54053aed0afaaa8f95297eff52a3d8b2512a; this draft remains for coordinated rollout approval.

This additive integration retains the deployed governance pin c8b8f89ecf29574d7e24f6c65cfd925aefbdd007. Canonical registration follows the reviewed fleet rollout in CultureBotAI/culturebotai-claw#432; the later pin update must use its actual main-reachable merge commit. Draft candidate refs are not release pins.

Closes #58. Closes #59. Closes #60. Closes #61. Coordinates with CultureBotAI/culturebotai-claw#426, CultureBotAI/culturebotai-claw#436 and CultureBotAI/culturebotai-claw#440.