feat: Integrate AIC as local MCP server

[nan-bit]

This PR replaces the direct Python binary execution with a standardized Model Context Protocol (MCP) server integration.

Key Changes

• AIC Server: Implemented a local MCP server locally in .
• New Tools: Exposes , , , and as MCP tools.
• Architecture: Updated to register the server.
• Documentation: Updated to reflect the new architecture.

Fixes

Closes #64

[nan-bit]

I've pushed a significant update to address issues found during testing:

1. Fix Tool Naming Mismatch: Renamed the server tools (e.g., index_repo → aic_index) to match the exact names expected by the prompt definitions in setup.toml and others. This fixes the tool call failures.
2. Fix Database Isolation: Updated gemini-extension.json to run the server in ${workspaceFolder}. This ensures the .aic dependency graph is created locally per project, preventing conflicts when using the extension across multiple repos.
3. Universal Indexing: Upgraded the indexer to support all text files. It now uses a UniversalSkeletonizer that falls back to raw content for non-Python files, making the tool useful for polyglot repos.
4. Respect Ignore Files: Added logic to respect .gitignore and .geminiignore patterns, ensuring we don't index node_modules or other build artifacts.

[nan-bit]

I have reconciled the branch with the latest upstream changes and finalized the
refactor of the AIC integration:

1. Resolved Merge Conflicts: Synchronized with upstream/main and bumped the
   extension version to 0.2.0.
2. Removed MCP Server Dependency: As part of the transition to a more portable
   architecture, the mcpServers configuration has been removed from
   gemini-extension.json.
3. Preserved Semantic Indexing: Updated the setup command logic to maintain the
   semantic indexing phase (aic_index) while ensuring it no longer relies on the MCP
   protocol for execution.

These changes ensure the extension remains lightweight and compatible with the latest
Conductor workflow.