Report QA coverage and scope: say which worksheets were not assessed

[roosch269]

The QA report says nothing about worksheets it never looked at. The schema describes 12; a COBie workbook can also carry Job, Resource, Spare, Impact, Assembly, Connection and Issue, and those are currently parsed by nothing, checked by nothing, and absent from the output. Coverage was never computed at all — the workbook knows its sheet names, the schema knows its own, and the difference was never taken.

Adds a Coverage and scope block to the QA view and the PDF cover. Against the example workbook in this repo it reads:

The active profile describes 12 of 19 recognised COBie worksheets present. Does not affect QA score.

Not assessed — recognised COBie worksheets this profile does not describe: Assembly — 0 rows; Connection — 0 rows; Impact — 0 rows; Issue — 0 rows; Job — 0 rows; Resource — 0 rows; Spare — 0 rows

Three categories, because a plain present-minus-described set isn't true — it reports a client's own Notes tab as COBie work that failed assessment:

• recognised COBie, not described by the profile → Not assessed
• not a COBie worksheet → Additional worksheet outside this profile
• described but no data rows → Present, empty

Instruction is excluded from the count: it's part of the COBie template but carries guidance, not project data, so calling it either "outside the profile" or "not assessed" would be wrong.

Two things deliberately not done the obvious way:

• the described set is the profile's full sheet list, not the stage-filtered one — a sheet excluded by the selected stage is described, not unassessed, and using stageSheets would make coverage change whenever the user switches stage
• iteration is per source workbook, not per logical facility — _qaLogicalFacilityRows() keeps one representative per facility identity, so a second file describing the same facility would disappear from the report

Coverage is descriptive, never evaluative. It lives in its own object and never touches qaFindings or qaRuleResults, so it cannot reach _qaScoreTally() and move a score or the donut. No severity colour, no icon.

Also fixes npm test on macOS. tests/schema-regression.js spawned bare python, which macOS hasn't shipped since 12.3, so the suite died at the first file with spawnSync python ENOENT before running anything. It now probes for a working interpreter; PYTHON= still overrides.

12 checks in tests/qa-coverage-regression.js, wired into npm test. Score invariance asserts every tallied value is byte-identical before and after coverage is computed and rendered. Three of them exist because running the built tool in a real browser against the example workbook found what the structural tests couldn't see — spacing in the rendered sentence, the filename repeated on every entry, and the Instruction misclassification.

Verified: four suites green with no PYTHON set, and two consecutive release builds byte-identical.

---

Targeted at dev so the call on main stays yours.

dev and main differ only by three merge commits with no content between them, so this diff is identical either way — 8 files, +447/-16.