feat: Activity status generation for long-running agent tasks

[aecs4u]

Summary

Generate human-readable activity status messages during long-running agent executions so users can see what the agent is doing (e.g., "Searching codebase...", "Analyzing 15 files...", "Writing test cases...").

Prior Art

• AutoGPT generates activity status updates during graph execution — each block emits a status string that's streamed to the UI in real-time. Users see "Running code block...", "Waiting for API response...", "Processing 3/5 items..." instead of a blank spinner.
• OpenClaw shows tool names during execution but lacks descriptive progress messages.

Motivation

RustyClaw's TUI currently shows streaming LLM responses, but during tool execution (especially long-running commands, web fetches, or multi-step operations), users see minimal feedback. Activity status messages improve UX by making the agent's work transparent.

Proposed Design

1. Each tool emits status events via the event streaming system (feat: Agent event streaming for real-time progress tracking #87):

   enum ActivityStatus {
       Starting { tool: String, description: String },
       Progress { tool: String, message: String, percent: Option<u8> },
       Completed { tool: String, summary: String, duration: Duration },
   }

2. TUI renders a status bar below the input area showing current activity
3. Gateway mode streams status events to WebSocket clients
4. Tool-specific status messages:
   • execute_command: "Running cargo test..."
   • web_fetch: "Fetching https://example.com (2.3 KB received)..."
   • search_files: "Searching 142 files for pattern..."
   • sessions_spawn: "Spawning sub-agent for code review..."

Acceptance Criteria

• Tools emit structured activity status events
• TUI displays status bar with current activity
• Gateway streams status events to WebSocket clients
• Status messages are descriptive and context-aware
• Progress percentages where applicable (file operations, multi-item tasks)

Related Issues

• feat: Agent event streaming for real-time progress tracking #87 (Agent event streaming — this extends it with activity semantics)
• feat: Streaming responses (block & tool level) #62 (Streaming responses)

[aecs4u]

Baseline shipped — Live status frames for thinking/tool execution/safety warnings shipped in e77e993. Remaining: progress percentage, streaming to web UI.