Improve codebase investigatior

Improvement of the codebase investigator involves multiple steps and consideration. Some things that must be covered/considered:

- [ ] Should we rename the agent to be more generic (less correlated to codebase) --> explore_context, etc.
- [ ] Evaluate strict structure and compare its effectiveness with more relaxed and unstructured returns. The current structure was optimized for gemini 2.5, but with gemini 3, we need to ensure this is the best version
- [ ] Evaluate system instructions and agent description to steer the main agent to more effectively call and utilize this exploration agent to preserve context and improve quality. This must take into consideration for what user behaviors and journeys do or don't need a special subagent call as well as the impact on token count.

Note: If we believe a general agent is sufficient and a context exploration agent is not necessary, that can also be a valid end state to this issue. If that is the case, remove the specialized subagent and update system instructions as needed.