fix(agents): render subagent dispatch rows regardless of SDK tool name

[aletc1]

Summary

• Top-level Agent/Task rows in the chat were rendering as a bare "Agent" label with no description, while the Tasks sidebar showed the correct summary. Root cause: the inline renderer hard-coded part.type === "tool-Task", but the CLI-parity built-in subagents shipped in feat(agents): ship CLI-parity built-in subagents to embedded SDK #36 can emit the dispatch tool as tool-Agent — so those runs fell through to a catch-all that shows only the bare tool name.
• Dispatch is now matched by shape (isSubagentDispatchType() covers both tool-Task and tool-Agent) at all three call sites in assistant-message-item.tsx, so AgentTaskTool consistently produces "Running Subagent — <description>" with nested child tools.
• Hardened the generic fallback to render <ToolName> <summary> via a shared summarizeToolInput() helper — extracted from tasks-widget.tsx so the two surfaces stay in sync forever.
• Bumped @anthropic-ai/claude-agent-sdk 0.2.118 → 0.2.119 and the bundled CLI binary 2.1.118 → 2.1.119 (released in lockstep).

Test plan

• bun install && bun run claude:download — pulls SDK 0.2.119 and binary 2.1.119.
• bun run dev — launch the Electron app.
• Send a prompt that triggers a subagent (e.g. "Design fix plan for …"). Confirm the inline chat row reads "Running Subagent — <description>" both while running and after completion, not just "Agent".
• Confirm the Tasks sidebar shows the same row with the same description (regression check on the shared helper).
• Run 3 subagent-triggering prompts back-to-back — label should be stable across all of them (regression test for the "sometimes works, sometimes doesn't" symptom).
• Invoke a tool not in the registry (e.g. an MCP tool) — fallback should now render <ToolName> <summary> instead of a bare name.

🤖 Generated with Claude Code