Preserve listing coverage before comparing growth and forecasting history

[byliu-labs]

Restoring permission to an unchanged folder can currently produce a growth alert: the earlier scan completed with a denied listing, but snapshots do not persist that coverage evidence. Legacy snapshots can also feed a confident forecast although their listing coverage is unknown.

Persist reported listing coverage and require two complete, comparable samples before calculating growth, journal attribution, time-slider deltas, or forecasts (including folder-budget projections). Live/saved comparisons return 409 COVERAGE_INCOMPARABLE; existing mount-boundary checks remain in place. Old history remains readable but cannot establish a baseline until newly measured complete samples exist.

This addresses reported errors only; listing failures an engine never reports remain outside this fix.

The tests use the shared waitFor hang guard. A pre-existing Windows failure in the identity stand-in is also corrected: expected paths now use the same native-parent canonical spelling as policy, rather than matching raw temporary-path strings. The same 5000 versus null failure occurs on unchanged main; an alias-spelling reproduction fails before the fixture correction and passes afterward. Production identity checks are unchanged.

Validation: all four coverage regressions fail on main 2c28ff5 before the fix. Real permission changes, a fresh Node-process reload, and real HTTP comparison endpoints are covered. The affected suites pass locally, the latest focused suites pass 15/15 with zero skips, and complete npm run typecheck passes all three configurations. The generated UI matches its sources. The unchanged full platform workflow passed on macOS, Windows, Linux and Linux under the Portuguese locale on our fork at exact commit 30191604faa4d9ce9d531eb8cc4d8bbbf700fb8e.

The complete fork matrix is green. This PR remains draft because upstream Actions still awaits first-time contributor approval.