Adds an optional reasoning effort override for the Advisor on the Coder Agents settings page, using the same model selector + effort slider that other agent overrides use.

Changes

• codersdk/coderd: AdvisorConfig gains an optional reasoning_effort. It is only accepted alongside a dedicated advisor model_config_id and is validated against that model's selectable efforts (400 with the selectable values otherwise). GET ignores stray legacy values without a model config.
• chatd: when an advisor model override resolves and an effort is configured, the advisor call uses that effort as its fixed default/max. If the advisor model is unavailable and the runtime falls back to the chat model, the override effort is not applied.
• UI: AdvisorSettings now uses the shared ModelSelector with its Effort slider. Selecting a model initializes the effort from the model's default; an explicitly unset effort stays empty so the advisor follows the model config's default.
• UI: ModelSelector gains an optional unsetLabel entry ("Use chat model" for the advisor) so the model can be unset without the Clear button that also resets usage limits; it stays usable when no enabled models exist.
• UI hardening from review: the submit path sanitizes a stored effort that is no longer selectable, drops it when the saved model is proven unavailable, and preserves it while model configs are loading or errored.
• Docs: advisor platform-controls page documents the new Reasoning effort setting.

Testing

• Unit: TestChatAdvisorConfig_* (coderd), advisor runtime/override tests (chatd), Storybook stories for the settings page.
• Dogfood UAT on a local instance with the chat-advisor experiment: selected a reasoning model, moved the slider to High, saved, verified API round-trip and reload persistence, cleared and verified the effort is dropped, plus API validation assertions (effort without model 400, invalid effort 400 with selectable values).

This PR was authored by Mux (an AI coding agent) working on Mike's behalf.