[WIP] Optimize SCC scheduler not_ready_deps tracking

[KevinRK29]

The not_ready_deps was populated with each direct dependency ID to do the two operations below:

• mark a dependency as completed by discarding it
• check if all dependences are done based on if it's empty

All of these could be done with a counter. The ids are never actually being used/inspected, just the size. So with a counter, we can just populate a counter with len(dep_sccs), and rather than discarding an id we can just decrement the counter, and we can check if the counter is zero for when all dependencies have been completed.

This is roughly a ~2% speed up on a warm run.

[github-actions]

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅