daemon: configurable LLM concurrency

New config field "llm_concurrency" (default 1) controls how many
concurrent model calls the daemon runs. Worker pool scales to match.

src/daemon.rs

@@ -283,14 +283,13 @@ struct DaemonStatus {
 
 pub fn run_daemon() -> Result<(), String> {
     let choir = Choir::new();
-    // Workers: 2 for long-running loops (scheduler, session-watcher),
-    // plus 1 for the actual LLM job (pool capacity is 1).
-    // Non-LLM jobs (decay, health) also need a worker, so 4 total.
-    let names: Vec<String> = (0..4).map(|i| format!("w{}", i)).collect();
+    let llm_concurrency = crate::config::get().llm_concurrency;
+    // Workers: 2 for long-running loops + llm_concurrency + 1 for non-LLM jobs
+    let n_workers = llm_concurrency + 3;
+    let names: Vec<String> = (0..n_workers).map(|i| format!("w{}", i)).collect();
     let _workers: Vec<_> = names.iter().map(|n| choir.add_worker(n)).collect();
 
-    // LLM API: 1 concurrent call to control token burn rate
-    let llm = ResourcePool::new("llm", 1);
+    let llm = ResourcePool::new("llm", llm_concurrency);
     llm.bind(&choir);
 
     // Recover last_daily from previous status file