Add real-time agent status streaming to UI

[dmfenton]

Implement comprehensive status updates from the agent to the UI:

Backend changes:

• Add new AgentStatus values: 'executing' and 'error'
• Add new message types: ThinkingDeltaMessage, CodeExecutionMessage,
  ErrorMessage, PieceCompleteMessage, IterationMessage
• Update agent.py with AgentCallbacks dataclass for event callbacks
• Stream thinking text as deltas (not accumulated) for efficiency
• Broadcast iteration start, code execution start/complete, and errors

Frontend changes:

• Update types.ts with new message types and AgentMessage metadata
• Handle all new message types in useCanvas.ts
• Add iteration and thinking state tracking
• Update MessageStream with styled bubbles for each message type:
  • Error messages with red accent and expandable details
  • Code execution with expandable stdout/stderr output
  • Piece complete with success styling
  • Iteration pills showing current/max progress
• Update StatusPill to handle 'executing' and 'error' states

This enables showing everything happening in real-time as the agent
thinks, executes code, and draws.