fix(importer): skip files with missing yEnc segments instead of importing them corrupt (#681)

[javi11]

What

Fixes #681. When a multi-segment file could not be size-normalized, parseFile swallowed the error and imported the file anyway using the NZB's raw yEnc-encoded byte counts. Those encoded sizes produce wrong segment offsets → corrupt media file.

Now: when normalization fails because a required segment is missing from all providers (errors.Is(err, nntppool.ErrArticleNotFound)), the file is skipped. The existing aggregation loop drops it and continues, so the rest of the release imports normally.

Why this scope

The reported trigger: the part-size probe fetches a single segment; if that article returns 430 on the provider used, normalization fails for the whole file and the corrupt fallback runs. The corruption only happens when we'd fall back to encoded NZB sizes for a file we can't verify.

The skip is intentionally scoped to not-found errors:

• Missing article (430) → skip (fixes the corruption).
• Present but non-yEnc / no usable part size → keep using the NZB-declared sizes (long-standing, non-corrupting behavior; importer integration battery relies on this).

To make detection uniform, the "known not found" branches in normalizeSegmentSizesWithYenc now wrap ErrArticleNotFound (NonRetryableError.Unwrap lets errors.Is traverse the fetch path too).

Changes

• internal/importer/parser/parser.go — scoped skip-on-not-found in parseFile; not-found branches in normalizeSegmentSizesWithYenc wrap ErrArticleNotFound.
• internal/importer/parser/parser_skip_test.go (new):
  • TestParseNzbSkipsFileThatFailsNormalization — a 430 on a needed segment skips that file; the healthy file is kept with normalized (decoded) sizes, not raw encoded values.
  • TestParseNzbKeepsFileWhenYencHeaderUnavailable — present-but-no-yEnc article is not skipped.
• internal/importer/parser/parser_parsenzb_test.go — serve realistic yEnc for healthy single-segment fixtures.

Reviewer notes

• The skip is per-parse and final. A 430 from a provider that actually has the article but isn't tried for the body fetch would now drop the file rather than corrupt it — the safe outcome. Cross-provider / multi-segment-probe recovery of such files was deliberately deferred (separate, larger change).

Verification

• go test ./internal/importer/parser/... -race ✅, go vet ✅
• go test ./internal/importer/ -race (integration battery) ✅
• make → exit 0, 0 failures ✅