Connectivity survey overcredits shared taxon CURIEs despite exact strain endpoints #1472

The source-only CF313 correction in #1471 exposes a measurement bug in tests/test_community_level_connectivity_credit.py. Its _names() survey unions preferred names, labels and IDs, so a precise CF313 source carrying the shared NCBITaxon:34072 also credits every other Variovorax strain. This falsely decreases solely-community-credited taxa by 28 rather than one. The production auditor and readiness ranker already use exact-name precedence; closed #762 fixed the ranker, not this separate test survey.

Acceptance: reuse the existing ranker's strain-aware resolver for the survey's non-community endpoints; retain the original metric definition and all bounds except the independently measured mixed-scope ceiling (124 to 125 after #1471). Add a shared-genus fixture proving one precise source connects one of 28 and ambiguous ID-only pairwise endpoints do not silently fan out. Capture both-root pre/post data and old/new algorithms separately, so changes in the metric are not misreported as biological changes. Existing #524 concerns participant warning policy and stays open. Production matching semantics must not change.

Dedup: 788 all-state issue titles/bodies searched for survey/census/credited_solely/_names, all eight matches read; ignored-hidden-inclusive local code and earlier record-specific search covered. #1344/#1360 are earlier census-bound updates, not this algorithm defect. Comments not exhaustive. Source-based self-adversarial finding, not independent approval.