fix: comprehensive agent tracking and cleanup improvements

This commit fixes all 4 reported agent tracking bugs and adds comprehensive test coverage:

**Bug Fixes:**
1. Agents stuck in "running" state after completion
   - Added SubagentStop hook parsing in wsl_bridge.rs
   - Emits claude:agent-end events when SubagentStop hooks detected
   - Includes 8 new Rust tests for hook parsing

2. Agents persisting after disconnect
   - Added clearConversation() call on disconnect in tauri.ts
   - Prevents agents from persisting across sessions

3. "Kill All" button doing nothing
   - Added markAllErrored() call in AgentMonitorPanel after interrupt
   - Updates UI state immediately after killing process

4. Badge persisting after closing tab
   - Added clearConversation() call in conversations.ts deleteConversation()
   - Properly cleans up agent tracking when tab is closed

**Test Coverage:**
- Added comprehensive agents.test.ts with 24 new tests
- Tests all store methods: addAgent, updateAgentId, endAgent, markAllErrored, clearCompleted, clearConversation, runningAgentCount
- Added 8 Rust tests for SubagentStart/Stop hook parsing
- All 387 frontend tests pass
- All 426 backend tests pass

**Documentation:**
- Updated CLAUDE.md with comprehensive testing guidelines
- Documents coverage goals, console mocking strategies, and E2E integration testing patterns

✨ This fix was implemented with help from Hikari~ 🌸

CLAUDE.md

@@ -52,6 +52,7 @@ All new features, fixes, and significant changes should include tests whenever p
 ### Mocking Strategies
 
 #### Console Mocking
+
 When testing code that intentionally logs errors (like error handling paths), mock console methods to prevent stderr output that makes tests appear flaky:
 
 ```typescript
@@ -70,6 +71,7 @@ it("handles errors gracefully", async () => {
 ```
 
 #### E2E Integration Testing for Cross-Platform Code
+
 For code that calls platform-specific system APIs (like Windows PowerShell or Linux notify-send), use helper functions that build the command structure without execution. This allows CI to verify cross-platform compatibility on Linux-only containers:
 
 ```rust
@@ -99,6 +101,7 @@ fn test_e2e_notify_send_command_structure() {
 ```
 
 This approach:
+
 - Verifies command structure, argument order, and escaping logic
 - Tests cross-platform code paths without requiring the target platform
 - Allows CI to catch regressions in Windows-specific code whilst running on Linux