feat(app): add model-switching to btw_app()

[gadenbuie]

Summary

Adds a model_choices parameter to btw_app() that lets users switch between AI providers and models from within the app's status bar.

The new parameter accepts four values:

• "auto" (default) — uses models from btw.md if the app was launched with one, otherwise falls back to the provider's model list
• "btw_md" — shows the named client entries from the btw.md config as model choices, enabling switching between providers mid-conversation
• "provider" — fetches available models from the active provider's API (previous behavior)
• "none" — no model selector shown

When switching between providers ("btw_md" mode), the server creates a new client from the target config, transfers the system prompt, tools, and conversation history, and replaces any <thinking> content blocks that are incompatible across providers. A toast notification confirms the switch.

Provider model choices are populated asynchronously after session start to avoid a startup race condition. The status bar's token and cost counters reset when the chat is cleared, and the provider badge appearance was polished.

Verification

# With a btw.md that has named client entries, e.g.:
# client:
#   claude: anthropic/claude-opus-4-5
#   gemini: google-gemini/gemini-2.5-pro
btw::btw_app()
# A model selector appears in the status bar with "claude" and "gemini" as options.
# Switching mid-conversation carries the history and system prompt to the new provider.

# Or, to browse all models from your provider:
btw::btw_app(model_choices = "provider")

# To disable the selector:
btw::btw_app(model_choices = "none")