Add deduplication: keep image closest to camera within 2s window

Within a burst sequence from the same station, the image where the athlete
is physically closest to the camera gives the most accurate passage timestamp.
Proximity is measured by bib bounding box area (larger = closer).

When a duplicate is detected:
- New image closer: update timestamp + image path, delete old image
- Existing image closer: discard new image

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

backend/profile_db.py

@@ -24,20 +24,21 @@ async def init_db(db: aiosqlite.Connection) -> None:
         );
 
         CREATE TABLE IF NOT EXISTS passages (
-            passage_id   TEXT PRIMARY KEY,
-            profile_id   TEXT REFERENCES athletes(profile_id),
-            bib_number   TEXT,
-            station      TEXT NOT NULL,
-            timestamp_utc TEXT NOT NULL,
-            gps_lat      REAL,
-            gps_lon      REAL,
-            gps_alt      REAL,
-            confidence   REAL,
-            id_method    TEXT,
-            source_image TEXT,
-            needs_review INTEGER NOT NULL DEFAULT 0,
-            review_note  TEXT,
-            created_at   TEXT NOT NULL DEFAULT (datetime('now'))
+            passage_id      TEXT PRIMARY KEY,
+            profile_id      TEXT REFERENCES athletes(profile_id),
+            bib_number      TEXT,
+            station         TEXT NOT NULL,
+            timestamp_utc   TEXT NOT NULL,
+            gps_lat         REAL,
+            gps_lon         REAL,
+            gps_alt         REAL,
+            confidence      REAL,
+            proximity_score REAL NOT NULL DEFAULT 0,
+            id_method       TEXT,
+            source_image    TEXT,
+            needs_review    INTEGER NOT NULL DEFAULT 0,
+            review_note     TEXT,
+            created_at      TEXT NOT NULL DEFAULT (datetime('now'))
         );
 
         CREATE INDEX IF NOT EXISTS idx_passages_profile ON passages(profile_id);