feat: Integrate Semantic Graphing & Rich Skeletons for Context Efficiency

Is your feature request related to a problem?
Yes. Currently, Conductor struggles to scale with "brownfield" projects. It reads full source files to gather context, which quickly consumes token windows and increases costs. This makes managing large, existing codebases costly.

Proposed Solution
Integrate a Semantic Dependency Graph and Rich Skeletonization into the Conductor workflow (based on aic).

Instead of ingesting raw code, the agent should operate on "Rich Skeletons"—lightweight representations of the code that preserve the API contract but discard implementation logic.

Technical Implementation
This introduces a local AST-based indexing layer:

Rich Skeletons: Compressed file representations containing only signatures, docstrings, and side-effects (e.g., RAISES, CALLS). This reduces context size by ~90%.

Dependency Graph: A local SQLite index that maps file relationships. This allows the agent to fetch a file's context and its immediate dependencies automatically, reducing hallucination.

Zero-Latency Analysis: Uses strict static analysis (AST) rather than LLM calls to build the index.

Associated PR
I have implemented this in PR #61. The changes cover:

Core: Porting the aic Skeletonizer and SQLite DB layer.

/conductor:setup: Builds the semantic graph at project initialization.

/conductor:newTrack & /conductor:implement: Updated to consume skeletons instead of raw files for significantly deeper context at lower token cost.

Start by reviewing PR #61 and the linked aic project, then trace the /conductor:setup, /conductor:newTrack, and /conductor:implement entry points described in the issue. Done means the semantic graph and rich skeleton workflow are integrated so these commands consume the indexed context instead of raw files.