fix: AI chat freeze, performance, code quality (#774)

[datlechin]

Closes #774

Summary

Fixes AI chat freezing when query results are included, plus 14 additional performance and code quality improvements.

Critical (freeze fix)

• Cap currentQuery to 2000 chars in system prompt (was embedding full editor text, potentially MB)
• Move buildSystemPrompt off @mainactor into Task.detached (was blocking main thread)
• Cap cell values in query results summary to 200 chars (large JSONB/TEXT columns)
• Pre-send size check: warns if total context exceeds 100KB

Performance

• Inline suggestion token batching (50ms intervals instead of per-token MainActor dispatch)
• Code block highlighting dedup (~190 lines of copy-paste reduced to shared function)
• Scroll-to-bottom with defaultScrollAnchor(.bottom) (macOS 14+)

Features

• maxOutputTokens configurable per provider (was hardcoded 4096/8192)
• Gemini fallback model list when API fetch fails
• Conversation file size cap (500KB, trims to last 50 messages)

Cleanup

• Non-actor-isolated overloads for AIPromptTemplates
• Deleted dead test code (CoordinatorShowAIChatTests)
• Distinct provider icons in settings
• Connection refresh on change in AI panel

Test plan

• Build passes
• Open large SQL file, enable include query/results, send AI message - no freeze
• Query table with large JSONB column - AI chat works without lag
• Stream long AI response - chat stays at bottom
• Inline suggestions on fast model - no UI jank
• Gemini with wrong endpoint - shows fallback model list