fix(search): close three P2 correctness gaps from code review

- vector: increment embedding_revision whenever delete_embedding is
  called so the HNSW cache is invalidated after image/folder deletions,
  not just after inserts

- hnsw_index: fix race in build_index — read the revision before
  fetching embeddings and again after the parallel_insert; retry the
  whole build if the revision advanced during construction, ensuring
  the cached index always reflects a consistent snapshot

- commands: replace the fixed 4× over-fetch in semantic_search_images
  with a progressive doubling loop; start at exactly `limit` candidates
  and double the batch (capped at 8192) until the page is filled or the
  vector table is exhausted, preventing both under- and over-fetching

src-tauri/src/vector.rs

@@ -33,6 +33,12 @@ pub fn migrate(conn: &Connection) -> Result<()> {
 #[allow(dead_code)]
 pub fn delete_embedding(conn: &Connection, image_id: i64) -> Result<()> {
     conn.execute("DELETE FROM image_vec WHERE image_id = ?1", [image_id])?;
+    // Advance the revision so any cached HNSW index is invalidated after deletions.
+    conn.execute(
+        "INSERT INTO app_kv (key, value) VALUES ('embedding_revision', 1)
+         ON CONFLICT(key) DO UPDATE SET value = value + 1",
+        [],
+    )?;
     Ok(())
 }