fix: improve AI tagging accuracy and loop detection

Commit notes

Issues fixed: 1. Essentia loop detection was too aggressive, incorrectly tagging one-shot samples (like bass hits) as loops 2. AI batch tagging was generating identical tags for different samples

Changes to Essentia analyzer (analyzer.py): - Increased minimum duration for loop detection from 0.5s to 1.0s - Added energy consistency check (one-shots decay, loops sustain) - Increased loop confidence thresholds (0.6 -> 0.75) - Added transient count check in classification (one-shots have 1-2 transients) - Classification now requires high loop confidence + multiple transients + sustained energy to classify as 'loop'

Changes to AI tagging (ai-tagging.ts): - Lowered duplicate tag threshold from 70% to 30% (triggers individual tagging fallback sooner for better uniqueness) - Improved audio signature with more distinctive features - Made unique traits description more granular (captures subtle differences) - Enhanced batch prompt with clearer instructions about filename hints and sample differentiation - Restructured sample descriptions to prominently show filename keywords and sample type (one-shot vs loop)

Co-authored-by: armin.naimi <[email redacted]>