Experimental: Avoid subprocess in preprocessing

[d7919]

When using pcpp we can avoid using subprocess to call the pcpp executable and simply construct a CmdPreprocessor object directly. This can save a little potential overhead of starting new processes.

[d7919]

I have a branch which further refactors this a little to use temporary files instead of redirecting stdout. This is then compatible with thread parallelism (but this doesn't seem to give any real speed up).

I also have a variant which takes the same approach but reinstates the subprocess calls. This seems to give a bit of a speed up (presumably because the subprocess calls aren't blocking the gil)