#2791: feat: fuzzy search for model selection dropdown

Feature Description

Add fuzzy search / filtering to the model selection UI so users can quickly find and switch models by typing partial names, provider prefixes, or abbreviations instead of scrolling through a long dropdown list.

Motivation

The model list in the WebUI continues to grow as new providers and model variants are added (OpenAI, Anthropic, Google, open-source models, custom providers, etc.). Currently, users must scroll through the full list to find the model they want, which is slow and error-prone — especially when:

• The list contains 50+ models across providers
• Users know the model name they want but not its exact position in the list
• Switching models frequently during a session (e.g., using a cheap model for simple tasks and a powerful one for complex ones)

This is a common UX pattern in modern developer tools — VS Code command palette, Obsidian file switcher, and Raycast all use fuzzy matching to great effect.

Proposed Solution

Add a text input at the top of the model dropdown that supports:

1. Substring matching — typing sonnet filters to claude-sonnet-4, claude-sonnet-3.5, etc.
2. Fuzzy matching — typing g4o matches gpt-4o, cs4 matches claude-sonnet-4
3. Provider prefix filtering — typing anthropic/ or openai/ filters to that provider
4. Keyboard navigation — arrow keys to move through filtered results, Enter to select
5. Highlight matched characters in the results so users can see why a model matched

A lightweight library like Fuse.js (~6KB) or a simple custom scorer would be sufficient — no heavy dependencies needed.

┌──────────────────────────────────────┐
│ 🔍 Search models...                  │
├──────────────────────────────────────┤
│ ✓ claude-sonnet-4          anthropic │
│   claude-sonnet-3.5        anthropic │
│   claude-sonnet-3.5-turbo  anthropic │
└──────────────────────────────────────┘

Alternatives Considered

• Provider submenus — groups models by provider but still requires scrolling within each group; does not help when you know the model name
• Recent models section — helpful but does not solve the general search problem; could complement fuzzy search
• Browser autocomplete — native autocomplete is inconsistent across browsers and not styleable

Scope / Effort Estimate

Small / Medium — touches the model selector component (likely in static/), adds a lightweight fuzzy matching utility. No backend changes needed since the model list is already available client-side.

Additional Context

• Related: #698 (model modals), #1240 (provider/model source of truth)
• VS Code, Raycast, and Slack all use this pattern for similar dropdown/palette UX
• Would also benefit mobile users where scrolling long lists is especially painful