Fix: tool events colliding across agents and status regression on replay

[hellop11]

Problem

test_directory_size_skips_symlinks in tests/test_live_view.py could fail when multiple agents were active in the same session because TuiLiveView tracked tool call/output events using a dictionary keyed only by call_id.

This caused two issues:

Tool events could collide across agents. If two agents reused the same call_id, one agent's event could overwrite the other's, causing incorrect or missing tool events in the timeline.
Replayed tool-call events could regress completed tools back to running. When _record_tool_call_data() was called again for an existing event, it always reset the status to running, even if _record_tool_output_data() had already marked the tool as completed or failed.

Fixes #660

Fix
Change the event lookup key from call_id to (agent_id, call_id) so tool events are isolated per agent.
Update _record_tool_call_data() so that, when an event already exists, it only refreshes mutable fields such as tool_name and args. The status and result fields are preserved and continue to be managed exclusively by _record_tool_output_data().

Changes
strix/interface/tui/live_view.py: key tool event lookup by (agent_id, call_id) instead of call_id.
strix/interface/tui/live_view.py: preserve existing status and result when replaying tool-call events.
tests/test_live_view.py: add three regression tests:
test_same_call_id_different_agents_stay_isolated
test_replayed_tool_call_does_not_regress_completed_status
test_failed_result_keeps_failed_status_after_replay

Before / After

Before: Tool events from different agents could overwrite each other when they shared the same call_id, and replayed tool-call events could incorrectly change a completed or failed tool back to running.

After: Tool events remain isolated between agents by using (agent_id, call_id) as the lookup key, and replaying a tool-call event updates only mutable metadata without altering the tool's final status or result.

[greptile-apps]

Greptile Summary

This PR fixes tool event tracking in the TUI live view. The main changes are:

• Tool events are keyed by both agent id and call id.
• Replayed tool-call events refresh name and args without resetting status or result.
• Live-view hydration and parsing helpers received stricter typing.
• Regression tests cover cross-agent call-id reuse and replayed status preservation.

Confidence Score: 5/5

This looks safe to merge.

• No blocking issues found in the changed code.

Important Files Changed

Filename	Overview
strix/interface/tui/live_view.py	Tool event lookup now separates agents and preserves completed or failed status during replay.
tests/test_live_view.py	Adds tests for call-id isolation across agents and replay status preservation.

_{Reviews (1): Last reviewed commit: "Fix tool events colliding across agents ..." | Re-trigger Greptile}