Refactor subagent and workflow domain models; migrate access tiers and events to domain module

- Moved `AccessTier` and `SubagentEvent` enums to `zesdex_domain::subagent`.
- Consolidated workflow-related types into `zesdex_domain::workflow`.
- Updated references across the codebase to use the new domain models.
- Refactored tool execution logic to utilize a new `ToolExecutor` trait.
- Enhanced `AgentTurnService` to handle tool calls and events more effectively.
- Adjusted API handlers and state management to align with new domain structure.

apps/infrastructure/src/subagent/auto/engine.rs

@@ -25,11 +25,9 @@ use zesdex_domain::core::ChatMessage;
 
 const REVIEW_AGENT_ID: &str = "auto-review";
 
-/// Spawn a background thread that reviews changes and auto-fixes issues.
+/// Spawn a background task that reviews changes and auto-fixes issues.
 ///
-/// Everything runs synchronously on the background thread — no tokio
-/// runtime is created, avoiding the nested-runtime panic from
-/// reqwest::blocking inside block_on in tokio >= 1.38.
+/// Runs asynchronously using tokio::spawn.
 #[instrument(skip(turn_events))]
 pub fn spawn_background_review(
     workspace_roots: Vec<PathBuf>,
@@ -41,7 +39,7 @@ pub fn spawn_background_review(
     let agent_id = REVIEW_AGENT_ID.to_string();
     let agent_name = "Auto-Review".to_string();
 
-    std::thread::spawn(move || {
+    tokio::spawn(async move {
         let root = match workspace_roots.first() {
             Some(r) => r.clone(),
             None => {
@@ -140,8 +138,6 @@ pub fn spawn_background_review(
         };
 
         // 5. Call LLM to review the diff and suggest fixes.
-        //    No tokio runtime needed — LlmClient uses reqwest::blocking
-        //    internally, which is fine on a plain thread.
         let system_msg = ChatMessage::system(
             "You are an auto-review subagent. Your ONLY job:\n\
              1. Review the git diff below for:\n\
@@ -171,8 +167,7 @@ pub fn spawn_background_review(
             "Review and fix this git diff:\n\n```diff\n{truncated_diff}\n```"
         ));
 
-        // This is a sync call — no tokio runtime required on this thread.
-        let response = run_llm_review(&client, &[system_msg, user_msg]);
+        let response = run_llm_review(&client, &[system_msg, user_msg]).await;
 
         let response_text = match response {
             Ok(text) => text,
@@ -232,10 +227,10 @@ pub fn spawn_background_review(
     });
 }
 
-/// Run the LLM review call synchronously using reqwest::blocking.
-fn run_llm_review(client: &LlmClient, messages: &[ChatMessage]) -> Result<String, String> {
-    // Direct LLM call — no tokio, no tool calls, just Q&A.
-    match client.chat_with_tools_non_streaming(messages, None, Some(1024), Some(0.3), None) {
+/// Run the LLM review call asynchronously using ProviderService.
+async fn run_llm_review(client: &LlmClient, messages: &[ChatMessage]) -> Result<String, String> {
+    use zesdex_application::ports::ProviderService;
+    match client.chat(messages, None, Some(1024), Some(0.3)).await {
         Ok((msg, _)) => Ok(msg.content.unwrap_or_default()),
         Err(e) => Err(e.to_string()),
     }