feat: runtime model switching via /model command

[bOsowski]

Is your feature request related to a problem?
The Claude model can only be set at startup via CLAUDE_MODEL config. Users should be able to easily switch models on the go, similar to the Claude TUI interface.

Describe the solution you'd like
A /model command that allows per-user runtime model switching:

• /model — show current model
• /model <name> — switch model (any valid Claude model name)
• /model default — reset to server/CLI default

Describe alternatives you've considered
Changing the CLAUDE_MODEL env var and restarting the bot — requires server access, causes downtime, and affects all users.

Additional context
The override is per-user (stored in context.user_data) and does not affect other users. Implementation threads model_override through the facade and SDK integration layers.

[RichardAtCT]

Tracked by PR #160 (runtime model switching with inline keyboard UI and per-model effort levels). PR needs rebase to resolve merge conflicts, then will be ready to merge. PR #139 (simpler implementation) has been closed in favor of #160.

[xiaotonng]

Per-user runtime model switching is the right shape for this — startup-only env vars don't survive long-running deployments. We hit the same in pikiclaw and ended up resolving model + reasoning effort per stream rather than per process: each session pulls from a runtime config that gets mutated by /model (and a dashboard picker), so different chats can run different models against the same agent CLI without restart. Listing live models per agent (Claude/Codex/Gemini) was the trickier piece — we probe each CLI on demand via a small list-models-live helper rather than hard-coding model names.

If the multi-tenant angle here interests you, we've got an open issue on user identity isolation that's the same threading-config-through-the-stack territory: xiaotonng/pikiclaw#3 — try npx pikiclaw@latest.

GitHub: https://github.com/xiaotonng/pikiclaw