Count evidence identities before computing cross-dimension coverage

When assessment dimensions share findings, adding their totals double counts evidence. Compute call-level coverage from unique finding identities, exclude explicitly not-applicable results from the applicable denominator, and retain not-assessable and missing/unavailable results separately. Partial identity sets should not yield a complete percentage. A useful regression sequence is partial detail → complete retry: merging normalized camelCase and transport snake_case fields can otherwise leave a stale completeness flag or override an explicit partial flag. This supports the prior distinction between bounded findings and complete success. Generalized method observed 2026-09-08 from fixture-only model/UI regressions and independent read-only review in COR-4086, source provenance: https://github.com/e3-solutions/ender-inference/pull/158 and commit d0d73d430ec6d80b7e86dc2fa00056a06118432c. No production call data or benchmark claims.