Separate agent for impact analysis

[kcramakrishna]

Despite 'reasoning' features of LLMs/genAI, this feature is not only expensive but theoretically can miss dependencies in existing code bases. We should have a separate agent for 'impact analysys'.
This agent can be called during the planning stage or to validate the proposed code changes (Probably some time before the code is written). This will be a specialised agent which will use the full power of the advanced code indexing utility we hope to integrate: #10

This should reduce

1. superficial edits and associated roll backs
2. agents painting themselves into a corner
3. agents going down the wrong or sub-optimal path
4. Triggering auomatic HITL (based on some criteria) when stable parts of legacy code is being changed.