[firestaerter3]

Fixes #1052

A clean health sweep marked a file healthy without looking at its persisted hole map, so a file that was zero-filled during playback went back to healthy while the holes were still there (details in the issue).

Changes

• When a sweep finds nothing missing, judgeValidation now reads the file's metadata again and checks known_holes. Holes that no local PAR2 patch covers keep the file degraded, with error type known_holes. This goes through the existing degraded handling, so PAR2 repair is queued as for any degraded file.
• The verdict is never failed. The acceptable-missing threshold and the repair flow are unchanged.
• Holes count regardless of age or provider fingerprint. Those only decide whether playback re-fetches before padding, and a STAT sweep cannot show that a missing body is back.
• The read happens at verdict time, not when the check is prepared, so a hole recorded while the sweep runs is not missed.
• If that read fails, or the metadata is gone, the result is inconclusive (pending, retried later, no retry used) instead of healthy.
• A file that is not eligible for hole classification (encrypted, non-video) still becomes healthy as before.
• The loop that skips patched segments moves into a helper shared with classifyHoles.
• content_verify_test.go and content_verify_report_test.go now give their checker a metadata service, since the clean path reads metadata.

Known limits

• A hole is never removed from known_holes when a later fetch of that article succeeds. So a file whose provider has refilled the article stays degraded until PAR2 patches it or the record is reset. Removing a hole on a successful fetch would be a separate change in nzbfilesystem.
• There is still a very short window between reading the hole map and writing the healthy result. A hole recorded exactly then can still be missed. Closing that fully needs a version check in the metadata layer, which seemed too much for this fix.

Testing

• New tests, which return file_healthy without the fix:
  • sampled and full sweeps through CheckFile and CheckFilesBatch;
  • a stale hole from another provider set;
  • a hole added between prepare and verdict;
  • unreadable and deleted metadata;
  • the worker in batch and direct mode.
• Unchanged behaviour is also covered: a patched hole and a file with no holes become healthy, and an ineligible file with a clean sweep stays healthy.
• go vet and go test -race ./internal/health/... ./internal/holes/... ./internal/metadata/... pass.
• Not linted, and the frontend was not built; no frontend files change.

[javi11]

@firestaerter3 test are failing can you fix it?