perf: concurrent worker with claim/execute split + graceful shutdown

- JobRepository::claim_next() — atomic SELECT FOR UPDATE SKIP LOCKED +
  UPDATE status=processing in one query, no duplicate claims
- ExecutePipelineHandler skips start() for already-claimed jobs
- Sweep spawns N concurrent tasks via JoinSet, claims are fast+sequential,
  execution is slow+concurrent
- Graceful shutdown: stop claiming, await all in-flight JoinSet tasks
- WORKER_CONCURRENCY env (default: CPU cores)
- DB_MAX_CONNECTIONS env (default: 20, was hardcoded 10)
- VolumeFileResolver impl for InMemoryFileStorage (test fix)

crates/adapters/postgres/src/db.rs

@@ -1,8 +1,12 @@
 pub type PgPool = sqlx::PgPool;
 
 pub async fn connect(url: &str) -> anyhow::Result<PgPool> {
+    let max_conn: u32 = std::env::var("DB_MAX_CONNECTIONS")
+        .ok()
+        .and_then(|v| v.parse().ok())
+        .unwrap_or(20);
     let pool = sqlx::postgres::PgPoolOptions::new()
-        .max_connections(10)
+        .max_connections(max_conn)
         .connect(url)
         .await?;
     Ok(pool)

crates/adapters/postgres/src/processing/mod.rs

@@ -145,7 +145,29 @@ impl JobRepository for PostgresJobRepository {
                     started_at, completed_at, error_message
              FROM jobs WHERE status = 'queued'
              ORDER BY priority DESC, created_at ASC
-             LIMIT 1",
+             LIMIT 1
+             FOR UPDATE SKIP LOCKED",
+        )
+        .fetch_optional(&self.pool)
+        .await
+        .map_pg()?;
+
+        Ok(row.map(Into::into))
+    }
+
+    async fn claim_next(&self) -> Result<Option<Job>, DomainError> {
+        let row = sqlx::query_as::<_, JobRow>(
+            "UPDATE jobs SET status = 'processing', started_at = NOW()
+             WHERE job_id = (
+                 SELECT job_id FROM jobs
+                 WHERE status = 'queued'
+                 ORDER BY priority DESC, created_at ASC
+                 LIMIT 1
+                 FOR UPDATE SKIP LOCKED
+             )
+             RETURNING job_id, job_type, target_asset_id, batch_id, status, priority,
+                       payload, result_data, retry_count, max_retries, created_at,
+                       started_at, completed_at, error_message",
         )
         .fetch_optional(&self.pool)
         .await