[FEATURE] Claude Desktop: show sub-agent activity (tool calls, commands, results) in the task panel and session history

### Preflight Checklist

- [x] I have searched existing requests and this feature hasn't been requested yet
- [x] This is a single feature request (not multiple features)

### Problem Statement

When Claude Desktop hands work to sub-agents, I have no way to check what they actually did: which commands ran, which files they read or changed, what came back, and what they finally answered. The panel shows a one-line summary while they run and "No output captured." once they finish (that fallback is filed separately as #88621). Reopening the session from history shows the same.

The information is already saved on disk, as a full transcript per sub-agent (`~/.claude/projects///subagents/agent-.jsonl`). The app just does not show it. For work where I have to answer for every action (infrastructure, production access) I need that record in the app, without digging through JSON files by hand.

### Proposed Solution

Show the sub-agent's transcript in its task panel entry: the tool calls in order (tool name plus the command, file path or prompt), each result collapsible, and the final answer. Keep it available when the session is reopened from history. Concretely, for agent-type tasks (`local_agent`, `local_workflow`, `remote_agent`, `in_process_teammate`):

1. Render the transcript from the task output file / subagents JSONL. Reusing the existing `local_bash` output-file reader (tail + size cap) would already be a big step.
2. In history, read the durable `subagents/agent-.jsonl` (not only the tmp `.output`), honouring `cleanupPeriodDays`.
3. An "Open transcript file" action (reveal in Finder / open path) as a fallback when rendering is not possible.

Related: #76669 (live tail of subagent output), #5974 (closed not-planned, TUI), #77510 (resumed-agent display bug), and the bug #88621.

### Alternative Solutions

- Minimal: show a link / path to the transcript file instead of "No output captured."
- Export action: "Copy sub-agent activity as Markdown" (tool calls + commands) for pasting into tickets.

### Priority

High - Significant impact on productivity

### Feature Category

Other

### Use Case Example

Operator delegates "audit IAM roles in account X" to 3 sub-agents, then must answer "which read-only AWS CLI commands were run, against which account, and what came back?" Today that takes:

```bash
jq -r 'select(.type=="assistant")|.message.content[]?|select(.type=="tool_use")|"\(.name)\t\(.input.command//.input.file_path//"")"' \
~/.claude/projects///subagents/agent-.jsonl
```

With this feature it is one click in the task panel, live and from history.

Start by tracing the existing local_bash output-file reader and the task panel/history entry points. Compare its tail and size-cap behavior with durable subagents/agent-.jsonl files under ~/.claude/projects///, including cleanupPeriodDays. Done means agent-type tasks show ordered tool calls, results, and final answers in the panel and after reopening, with a file-opening fallback.

由索引模型根据 Issue 内容生成。