feat: Phase 3 API (Identity, Files, Candidates) and pre_chunks migration

migrations/017_create_pre_chunks.sql

@@ -0,0 +1,47 @@
+-- ============================================================================
+-- Migration 017: Create pre_chunks table (Processor Output)
+-- ============================================================================
+-- Purpose:
+--   1. Move raw processor outputs (YOLO frames, Face detections, etc.)
+--      from 'chunks' table to a dedicated 'pre_chunks' table.
+--   2. Support coordinate_type (frame for video, text for audio, etc.)
+--      to allow future expansion for PDF/Audio files.
+--   3. Support Identity linking directly on pre_chunks (Face -> Identity).
+-- ============================================================================
+
+-- 0. Clean up existing conflicting table (if any)
+DROP TABLE IF EXISTS pre_chunks CASCADE;
+
+-- 1. Create pre_chunks table
+CREATE TABLE pre_chunks (
+    id                  BIGSERIAL PRIMARY KEY,
+    file_uuid           UUID NOT NULL,
+    processor_type      VARCHAR(32) NOT NULL,      -- 'yolo', 'face', 'asr', 'ocr', 'pose'...
+    
+    -- Coordinate system (supports Video, Audio, Text...)
+    coordinate_type     VARCHAR(20) DEFAULT 'frame', -- 'frame', 'time', 'page'
+    coordinate_index    BIGINT NOT NULL,           -- Frame number, or paragraph index
+    
+    timestamp           FLOAT,                     -- Time in seconds
+    data                JSONB NOT NULL,            -- Raw processor output (objects, bboxes, etc.)
+    
+    -- Identity linkage (Face -> Identity, or Speaker -> Identity)
+    -- If NULL, this Face/Speaker is a "Candidate"
+    -- Note: FK removed temporarily due to schema migration in progress
+    identity_id         UUID, 
+    
+    confidence          FLOAT,                     -- Match confidence
+    
+    created_at          TIMESTAMPTZ DEFAULT NOW()
+);
+
+-- 2. Indexes
+CREATE INDEX idx_pre_chunks_file ON pre_chunks(file_uuid);
+CREATE INDEX idx_pre_chunks_processor ON pre_chunks(processor_type);
+CREATE INDEX idx_pre_chunks_identity ON pre_chunks(identity_id);
+CREATE INDEX idx_pre_chunks_coord ON pre_chunks(file_uuid, processor_type, coordinate_index);
+
+-- 3. Comment
+COMMENT ON TABLE pre_chunks IS 'Raw output from Processors (Frames, Segments). Candidates are rows where identity_id IS NULL.';
+COMMENT ON COLUMN pre_chunks.coordinate_type IS 'Coordinate unit: frame (Video), time (Audio), page (PDF)...';
+