fix: surface doc parse failures instead of empty success

[fpt]

Summary

Fixes the two robustness gaps found while verifying the dq-based parsers against live pkg.go.dev. The same pattern existed in all three doc readers (godoc, rustdoc, pydoc).

The problems

In Read*Paged and SearchWithin*:

1. Silent empty on parse failure — the parser's matched flag was discarded (_, document = parse...). When all parsers missed (e.g. the site renamed a CSS class), the reader returned "" with a nil error, indistinguishable from "this package has no docs."
2. Caching failures — docCache.Set ran even for an empty document, poisoning the cache for the 30-minute TTL on a transient miss or markup drift.

These matter because the parsers are CSS-class-coupled and the golden tests run against saved HTML, so they can't catch live drift — the failure mode was silent empty output.

The fix

Read*Paged / SearchWithin* now check the parse result and return ErrNotFound when nothing matched (or the parsed document is blank), and only cache non-empty parses. Applied consistently across godoc.go, rustdoc.go, pydoc.go (6 call sites).

matched, parsed := parseDocsRsDocument(doc)
if !matched || strings.TrimSpace(parsed) == "" {
    return "", 0, false, ErrNotFound
}
document = parsed
docCache.Set(cacheKey, document, cache.DefaultExpiration)

(godoc keeps its parseDocument → parseReadme fallback; the guard runs after both.)

Verification

• go build, go vet, full go test ✅
• make lint → 0 issues
• Golden tests unaffected (they exercise parse* directly, not the readers)

🤖 Generated with Claude Code