feat: Configurable minimum-score threshold (+ fix for ranking authors with initials)

[NichCodes]

Fix torrent ranking for authors with initials; make the minimum-score threshold configurable

Problem

1. The ranking algorithm's author-presence gate could not reconcile compacted
   initials ("JRR Tolkien") with spaced initials ("J R R Tolkien"). Because that gate
   hard-zeros the entire 60-point title/author match when the author isn't found,
   correct, well-seeded releases were scored far below the accept threshold and
   rejected — leaving requests stuck in awaiting_search. It affected any
   initials-style author (JD Robb / J D Robb, George RR Martin / George R R Martin,
   W E B Griffin, etc.).
2. The automatic-search accept threshold was hard-coded, so the only way to make
   it stricter or more lenient was a code change. Different setups need different
   bars (a single high-quality indexer vs. many noisy public ones).

Fix

• Add an initials-normalization step (collapseInitials) that joins runs of
  single-letter tokens, applied in checkAuthorPresenceWithParsed and the
  Stage-3 author substring scoring so spaced and compacted initials compare
  equal.

• Make the accept threshold admin-configurable instead of a hard-coded number.
  Two optional sliders on Settings → Indexers (styled like the Indexer Flag
  Configuration section, marked Optional):

  • Audiobooks → config key indexer.min_quality_score
  • E-books → config key indexer.min_quality_score_ebook

  Each ranges 0–100, default 50, with inline help explaining the trade-offs
  (too high rejects everything and stalls requests in awaiting_search; too low
  grabs weaker matches) and a warning when set to 0. The title/author match
  gate (matchScore > 0) still applies independently, so even at 0 a wrong
  title/author is rejected. Manual/interactive searches remain unfiltered.
  The search processors read these keys at runtime (clamped 0–100; missing/invalid
  → 50).

• For e-books, the same change also adds the title/author match gate
  (matchScore > 0) the audiobook path already had, so a wrong-author release can
  no longer slip through on size/seeder/format score alone.

• The clamp/default logic is centralized in one helper
  (parseMinQualityScore / DEFAULT_MIN_QUALITY_SCORE) shared by all four read
  sites, since the settings GET, the indexer-options route, and both processors
  must interpret the keys identically.

Tests

• Regression coverage for ranking: the initials variants and a
  wrong-author negative to confirm the gate still rejects mismatches.
• Processor + settings coverage: search-indexers / search-ebook honor the
  configured threshold (default 50) and the match gate; indexer-options persists
  and validates the new keys (integer 0–100); settings save sends them.

Scope

• src/lib/utils/ranking-algorithm.ts — initials normalization
• src/lib/processors/search-indexers.processor.ts,
  src/lib/processors/search-ebook.processor.ts — read configurable threshold
• src/lib/utils/min-quality-score.ts — shared parse/clamp helper + default
• src/components/admin/MinScoreSlider.tsx — new slider component
• src/app/admin/settings/tabs/IndexersTab/IndexersTab.tsx,
  src/app/admin/settings/lib/{types,helpers}.ts — UI + wiring
• src/app/api/admin/settings/route.ts,
  src/app/api/admin/settings/indexer-options/route.ts — load/save/validate
• Docs: documentation/settings-pages.md, documentation/phase3/ranking-algorithm.md,
  documentation/TABLEOFCONTENTS.md
• Ranking test file + processor/helpers tests