Rethink native-tools UI: default-on, demote disable to advanced settings

[dsfaccini]

David's AICA here:

Problem

The chat toolbar surfaces the provider-native tools (web search, code execution, image generation) as prominent per-tool toggles next to the model picker (see the switches wired in src/Chat.tsx, and the tools attached in agent/chatbot/server.py).

When this UI was designed, exposing each tool as an explicit opt-in made sense. LLM development and user expectations have moved on: people now generally expect an assistant to already have web search / code execution / etc. available. Making each one a primary, always-visible control reads as dated and adds noise to the main input area, when most users just want the tools on.

Proposal

• Enable the native tools by default.
• Move the disable controls out of the primary toolbar into an advanced / settings surface, for the minority who explicitly want to turn a tool off.
• Keep the main toolbar focused on the things users actually change per-message (model, and -- per feat: reasoning-effort dropdown in the chat toolbar #28 -- reasoning effort).

Open questions

• Where should "advanced settings" live (a popover off the toolbar, a settings dialog, a per-conversation menu)?
• Should the default-on set be per-model (some models/providers don't support all native tools)?
• Do we persist per-user overrides (localStorage, like model/effort) or reset to defaults each session?

Not urgent; filing to track the rethink.