Improve performance of usage command

[lars-reimann]

Currently, there is a lot of file I/O happening. After a single program is processed, its usages are written to disk. Once everything is processed, the usages are aggregated. This ensure that in case the program crashes we do not need to start from scratch and allows for parallelization. However, it's still really inefficient.

Alternative solution:
Have workers just return their results and continuously aggregate them. Ensure the program does not crash 🙃.

Additional Context

• Possible memory leak caused by astroid: Continuously increasing memory usage when pylint is run via its API pylint-dev/astroid#792