feat: add GitHub Copilot support with rewritten inline suggestion architecture

[datlechin]

Summary

Closes #754

Adds GitHub Copilot support to TablePro with a complete rewrite of the inline suggestion architecture and a new AI provider registry pattern.

Inline Suggestion Architecture Rewrite

• Decompose the 464-line InlineSuggestionManager into a clean, protocol-based system
• InlineSuggestionSource protocol for pluggable suggestion sources
• GhostTextRenderer for CATextLayer ghost text display
• AIChatInlineSource for existing AI provider path (Claude, OpenAI, etc.)
• Dynamic source resolution via closure (settings changes take effect immediately)

Generic LSP Client

• Core/LSP/ with LSPTransport (stdio JSON-RPC 2.0), LSPClient (typed methods), LSPTypes, LSPDocumentManager
• Reusable for any language server (SQL language servers, linters, etc.)

GitHub Copilot Integration

• CopilotService: language server lifecycle, OAuth device flow auth, exponential backoff restart
• CopilotBinaryManager: download native binary from npm with SHA1 integrity verification
• CopilotInlineSource: inline completions via textDocument/inlineCompletion with proper ghost text rendering and replacement range handling
• CopilotDocumentSync: editor text changes synced to LSP with server-restart awareness
• CopilotSchemaContext: database schema preamble (table names, columns, types) prepended to document for schema-aware suggestions
• CopilotChatProvider: chat via conversation/create + conversation/turn LSP protocol with streaming

AI Provider Registry

• AIProviderDescriptor + AIProviderRegistry + AIProviderRegistration for extensible provider registration
• AIProviderFactory tries registry first, hardcoded switch as fallback
• Adding a new provider (e.g., Codeium) = 1 file + register call

Unified Settings

• Single InlineSuggestionProvider picker in Editor settings (Off / GitHub Copilot / AI Provider)
• Copilot settings tab with sign-in flow, status indicator, telemetry toggle
• Copilot Chat toggle in AI settings with feature routing support

Architecture

AIProviderRegistry
    ├── claude    → AnthropicProvider (chat)
    ├── openAI    → OpenAICompatibleProvider (chat)
    ├── copilot   → CopilotChatProvider (chat via LSP)
    └── [future]  → CodeiumProvider, SupermavenProvider, ...

InlineSuggestionManager (orchestrator)
    ├── GhostTextRenderer (CATextLayer)
    └── InlineSuggestionSource (protocol)
            ├── AIChatInlineSource (AI providers)
            └── CopilotInlineSource (Copilot LSP)
                    └── CopilotService → LSPClient → LSPTransport → Process

Files (36 changed)

New (18):

• Core/LSP/ — LSPTypes, LSPTransport, LSPClient, LSPDocumentManager
• Core/AI/InlineSuggestion/ — InlineSuggestionSource, GhostTextRenderer, AIChatInlineSource, InlineSuggestionManager
• Core/AI/Copilot/ — CopilotService, CopilotAuthManager, CopilotBinaryManager, CopilotDocumentSync, CopilotInlineSource, CopilotSchemaContext, CopilotChatProvider
• Core/AI/Registry/ — AIProviderDescriptor, AIProviderRegistry, AIProviderRegistration
• Models/AI/CopilotSettings.swift, Views/Settings/CopilotSettingsView.swift

Modified (8): AIModels, AIProviderFactory, SQLEditorCoordinator, SQLEditorView, EditorSettings, EditorSettingsView, AISettingsView, AppSettingsManager, AppSettingsStorage, SettingsView, TableProApp

Deleted (1): Core/AI/InlineSuggestionManager.swift (replaced by decomposed files)

Test plan

• Build succeeds
• InlineSuggestionManagerFocusTests pass (12/12)
• AISettingsTests pass (3/3)
• Copilot server starts on app launch when enabled
• Copilot auto-detects existing authentication
• Copilot inline suggestions show real table/column names from schema
• Ghost text renders correctly, Tab accepts, Escape dismisses both popup and ghost text
• Settings > Editor > Inline Suggestions picker works (Off / Copilot / AI)
• Settings > Copilot > sign-in/out, status indicator
• Settings > AI > Copilot Chat toggle, feature routing
• Copilot Chat streaming via conversation protocol
• Verify no regressions in Vim mode + inline suggestions