Currently we only perform test discovery on the whole project. This is unnecessary when discovery happens in response to a file being created, updated, or deleted. We should be able to perform discovery on a per-file basis in those cases. Doing so will be more efficient and allow us to be more granular in how we present test status.

This is effectively blocked by a uniform test discovery API for the different test frameworks (e.g. like we're doing in #3911 for pytest).