[FEATURE] Let MassGen decide when semantic search is necessary

[ncrispino]

Feature Description

We do need semantic search if a filesystem is large, else we will waste much time and money reading all files. Instead of imposing semantic search on MassGen, we should be able to have the agents handle the context smartly themselves. Based on MAS-77, models should be smart enough to know when the context is filling up.

Basically, we should write a skill that gives us semantic search options like semtools for the model to use. We also need some way to tell the model when the skill should be used. I am thinking we find some good terminal commands (like we have for rg, ast-grep) that can tell us the size of the workspaces and if reading the context would be practical. I think it's entirely reasonable for the model to be able to e.g., see a 1million context repo, know it has a context of 250k and say hey, I need to do indexing and smarter searching. Then it can use the semantic search skill to easily index the workspace and ask questions more easily.

This is useful bc there's overhead to semantic search and a lot of times it may not be necessary.