fix(runner): restore tool error reporting on ToolCallEndEvent (#1100)

## Summary

- Restores `is_error` forwarding from `ToolResultBlock` to
`ToolCallEndEvent` in the AG-UI handler
- PR #613 (Feb 18) refactored to event-driven architecture but dropped
this field, breaking Langfuse tool error tracking since Feb 24
- 3-line fix matching the old `adapter.py` behavior (lines 836-846)

## Root Cause

`handle_tool_result_block()` in `handlers.py` reads `is_error` from the
Claude SDK's `ToolResultBlock` but never forwards it to
`ToolCallEndEvent`. The observability layer (`track_agui_event` in
`observability.py`) checks `getattr(event, "error", None)` on
`TOOL_CALL_END` events and always gets `None`, so all tool spans are
tagged `level=DEFAULT` instead of `level=ERROR`.

## Impact

- Langfuse tool error tracking broken for ~5 weeks (since Feb 24)
- Cockpit dashboard correction rate dropped from ~8% to 0.01%
- Session metrics `tool_failures_total` always reports 0
- `tool_log_correction` MCP tool still works (separate path) but only
captures explicit human corrections, not tool execution failures

## Test plan

- [ ] Verify `ToolCallEndEvent` accepts `result` and `error` fields
(confirmed via test fixtures in `conftest.py`)
- [ ] Run `pytest components/runners/ambient-runner/tests/`
- [ ] After deploy: verify Langfuse observations show `level=ERROR` on
tool failures
- [ ] Cockpit correction rate dashboard should show non-zero values
again

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: 2 people

components/runners/ambient-runner/ag_ui_claude_sdk/handlers.py

@@ -154,6 +154,7 @@ async def handle_tool_result_block(
     """
     tool_use_id = getattr(block, "tool_use_id", None)
     content = getattr(block, "content", None)
+    is_error = getattr(block, "is_error", None)
 
     # Parse tool result content for frontend rendering
     # Claude SDK tools return: [{"type": "text", "text": "{json_data}"}]
@@ -191,6 +192,8 @@ async def handle_tool_result_block(
             thread_id=thread_id,
             run_id=run_id,
             tool_call_id=tool_use_id,
+            result=result_str if not is_error else None,
+            error=result_str if is_error else None,
             timestamp=now_ms(),
         )