Daemon: --agent flag on chat and code commands

[ludusrusso]

Parent PRD

#34

What to build

Add --agent flag to wildgecu chat and wildgecu code commands. When provided, the daemon creates a session scoped to the named agent's Home, SOUL, model, and memory. Includes memory curation scoped to the agent after session close.

End-to-end: wildgecu chat --agent=researcher connects to the daemon, creates a session using the researcher's SOUL and model, and after the conversation the researcher's MEMORY.md is updated.

Acceptance criteria

• --agent flag added to chat and code commands
• ChatRequest IPC protocol gains an Agent field
• Daemon creates sessions using the named agent's Home (SOUL, MEMORY, model)
• agent.Prepare() / agent.PrepareCode() work with agent-scoped Home
• Memory curation (agent.Finalize) writes to the agent's MEMORY.md, not the main agent's
• When --agent is omitted, existing behavior is preserved (backward compatible)
• Error when agent name doesn't exist

Blocked by

• Blocked by Home.Agent: directory scaffolding for static subagents #41 (Home.Agent directory scaffolding)
• Blocked by Agent config.yaml: parsing and model resolution #42 (agent config.yaml parsing)
• Blocked by CLI: wildgecu agent add <name> bootstrap #43 (agent add — need an agent to test with)

User stories addressed

• User story 4 (chat with specific agent)
• User story 5 (code mode with specific agent)
• User story 8 (per-agent memory)
• User story 17 (per-agent memory curation)