Fix matching regressions, cache guard, and multi-shot algorithm for beat 15

- config.toml: revert scoreable_luma/contrast thresholds to 24/58/24 (lowering
  them let cross-fade blend frames contaminate content-validation templates,
  dropping scores below provisional_content_threshold)
- src/cv/global_scan.py: _is_dark_reference_frame now requires contrast<30 so
  genuine dark silhouette frames are not rejected as scoreable; two-path
  _is_scoreable_reference_frame separates standard vs fade-content scoring
- cli.py: _keeps_cached_match() guard prevents a weaker single-span rematch
  from overwriting a better multi-segment provisional cache entry
- cli.py: _fade_content_shots() restricted to between-island gaps only—
  pre-island black leaders were incorrectly emitted as matchable shots
- cli.py: island[0] of _match_unmatched_visual_segments() now uses no
  continuity seed so an insert cut at the start of a multi-shot beat is not
  forced toward the previous beat's scene
- scripts/generate_cutter_report.py: fix ffmpeg concat demuxer on Windows—
  use part.absolute().as_posix() so paths in the concat txt are absolute and
  not double-resolved relative to the concat file's directory

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

config.toml

@@ -72,12 +72,12 @@ match_threshold       = 0.65
 
 # Store/report lower-confidence automatic candidates for visual review instead
 # of dropping them as "NO MATCH". Confirmed exports can still use match_threshold.
-provisional_match_threshold = 0.43
+provisional_match_threshold = 0.35
 
 # Lower gate for entering temporal multi-frame refinement. The final decision
 # still uses sequence/span scoring; this only avoids rejecting real matches
 # because one midpoint frame is weak.
-coarse_candidate_threshold = 0.50
+coarse_candidate_threshold = 0.40
 
 # Candidate ranking weights. Duration coverage matters when the same visual
 # shot appears multiple times: prefer the occurrence that can cover the beat.
@@ -103,7 +103,7 @@ refine_step_seconds   = 0.04  # ≈ 1 frame at 25 fps
 content_align_window_seconds = 0.48
 content_align_sample_step_s  = 0.28
 content_validation_weight    = 0.35
-provisional_content_threshold = 0.42
+provisional_content_threshold = 0.30
 
 # When several adjacent frame offsets score almost the same, prefer the earlier
 # one. This avoids matches that are visually correct but start a few frames late.