[FEATURE]: Sub-agent abort recovery — show task_id and allow retry

[Ithril-Laydec]

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

Problem
When a sub-agent is aborted (connection drop, crash, Ctrl+C, or system shutdown), the TUI shows "Tool execution aborted" with no actionable information. The user has no way to:

1. Identify which child session was running (no task_id / session ID visible)
2. Resume the aborted sub-agent (no retry action available)
   The child session still exists in the database with its full message history, and the Task tool already supports resumption via task_id — but the UX provides no path to use it.
   This is especially painful for long-running sub-agents (complex implementations, multi-file refactors) where losing 5-10 minutes of work to a transient connection issue is a real problem.
   Current behavior
   ┃ General Task — Implement authentication module
   ┃ Tool execution aborted
   No session ID. No retry option. The user must start over or manually dig through session data.
   Proposed solution
3. Display task_id on aborted Task tool parts (quick win, ~5 lines)
   The data is already available — metadata.sessionId is preserved by the abort handler in processor.ts (it spreads existing metadata and adds interrupted: true). The Task component in session/index.tsx just doesn't render it in the error state.
   ┃ General Task — Implement authentication module
   ┃ └ task_id: ses_0191a3b4c5d6AbCdEfGhIj
   ┃ Tool execution aborted
   This alone makes manual recovery possible (user can tell the parent agent to resume with that ID).
4. "Retry" action on aborted Task tool parts (the real feature)
   A keybind (e.g., r when focused on an aborted tool part) that re-invokes the Task tool call with the same parameters plus task_id injected automatically. The child session continues where it left off and the result flows back to the parent naturally.
   This approach:

• Uses the existing task_id resumption mechanism in tool/task.ts
• Maintains the parent ↔ child relationship (result returns to parent)
• Doesn't depend on LLM interpretation of a text prompt
• Could be extended to a generic "retry aborted tool" mechanism in the future
  Key implementation details
• Abort preserves metadata (processor.ts ~L506): metadata: { ...metadata, interrupted: true } — so sessionId is always available
• Task tool supports resumption (tool/task.ts ~L63): when task_id is provided and the session exists, it reuses it with full history
• Click-to-navigate already works: the Task component's onClick navigates to the child session via metadata.sessionId — the data path is proven
  Related issues
• TaskTool returns no diagnostic context when subagent result text is empty or missing #24447 — TaskTool returns no diagnostic context when subagent result is empty (proposes showing child session ID)
• Opencode hangs indefinitely if exited during waiting for subagents to finish #24391 — OpenCode hangs if exited during subagent execution (the abort scenario)
• Main & Sub-agents Randomly Freeze Indefinitely: Frontend Permanently Shows "thinking" with No Errors, While Actual LLM Inference Has Already Terminated Prematurely #24342 — Sub-agents randomly freeze with no recovery path

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• [FEATURE]: Recovery path for errored subagent sessions (composer lockout blocks manual continuation after provider failures) #21525: [FEATURE]: Recovery path for errored subagent sessions — covers the same core need of a retry/resume action for halted/aborted subagent sessions (including an explicit "Retry" button on the error banner, Option B in that issue). The abort scenario described here overlaps significantly with what's proposed there.