Consolidate memory RPC in tools/memory.rs

- Move memory_rpc(), socket_path(), SocketConn from mcp_server.rs
- Convert remaining callers to typed async API:
  - defs.rs: organize placeholder, run_agent query
  - cli/agent.rs: query resolution (now async)
  - mind/identity.rs: Store context loading
- Re-export socket_path/memory_rpc from mcp_server for compatibility

All external memory access now goes through tools/memory.rs typed API.

Co-Authored-By: Proof of Concept <poc@bcachefs.org>

src/cli/agent.rs

@@ -1,8 +1,9 @@
 // cli/agent.rs — agent subcommand handlers
 
+use crate::agent::tools::memory;
 use crate::store;
 
-pub fn cmd_run_agent(agent: &str, count: usize, target: &[String], query: Option<&str>, dry_run: bool, _local: bool, state_dir: Option<&str>) -> Result<(), String> {
+pub async fn cmd_run_agent(agent: &str, count: usize, target: &[String], query: Option<&str>, dry_run: bool, _local: bool, state_dir: Option<&str>) -> Result<(), String> {
     // Mark as agent so tool calls (e.g. poc-memory render) don't
     // pollute the user's seen set as a side effect
     // SAFETY: single-threaded at this point (CLI startup, before any agent work)
@@ -22,11 +23,10 @@ pub fn cmd_run_agent(agent: &str, count: usize, target: &[String], query: Option
     let resolved_targets: Vec<String> = if !target.is_empty() {
         target.to_vec()
     } else if let Some(q) = query {
-        // Use RPC to resolve query
-        let result = crate::mcp_server::memory_rpc(
-            "memory_query",
-            serde_json::json!({"query": format!("{} | limit:{}", q, count)}),
-        ).map_err(|e| e.to_string())?;
+        // Resolve query via typed API
+        let q_str = format!("{} | limit:{}", q, count);
+        let result = memory::memory_query(None, &q_str, None).await
+            .map_err(|e| e.to_string())?;
         let keys: Vec<String> = result.lines()
             .filter(|l| !l.is_empty() && *l != "no results")
             .map(|s| s.to_string())