Reasoning controls and shortcut-based context capture

This release gives PMs more control over both model reasoning and page context capture. You can now choose reasoning depth directly in the side panel model picker for harder tasks, and add highlighted text to context with a deliberate keyboard shortcut instead of relying on focus-based behavior.

The result is a more predictable workflow when you are collecting quotes from customer calls, competitor pages, docs, and internal notes: capture only what you want, when you want it, and control how deeply the model should think before responding.

What's new

• Reasoning depth controls in the side panel model popup with low, medium, and high options.
• Model catalog capability flags (supports_thinking) so reasoning controls only appear for compatible models.
• Provider-specific reasoning settings for OpenAI, Anthropic, and Google so selected depth is applied to runtime calls.
• Improved Gemini reasoning behavior by requesting thought content (includeThoughts) so supported reasoning output can appear in the UI.
• Persistent reasoning-level preference so your selected depth carries across sessions.
• Native highlighted-text shortcut capture (Ctrl+Shift+Y / Command+Shift+Y) in Chrome and Firefox.
• Focus-based selected-text auto-attach removed in favor of explicit shortcut-triggered capture.
• Clearer in-panel capture feedback and duplicate-selection protection for repeated url + exact text captures.