perf: Concurrent & optional pre-processing

[arvid-norlander]

This implements a pair of workarounds for #350.

• One is to use multi-core (most useful work working on a single target)
• One is to entirely skip using the preprocessor (even faster).

Another option could be to instead add the ability to call out to the pre-processor from a hermetic toolchain. However, we don't use conditional includes, so these workarounds works for me.

[martis42]

I do not plan to integrate the "no preprocessing at all" part. As said in the discussion in #350, I want at least to handle commented code. A different implementation of experimental_no_preprocessor, which ignores commented code but does nothing else regarding preprocessing, just shipped with release 0.9.0.

Regarding the concurrent processing I am not yet sure. Bazel expects one action not to consume more compute power than a single core of your CPU can provide. The multi_core option would violate this and also not work well in remote execution scenarios. So, this would be an option specialized on analyzing a few very complex targets.

That much said, I am not outright rejecting the idea. But could you please first do a new test of DWYU based on the 0.9.0 release which includes your contribution #352 and the experimental_no_preprocessor option. I hope, those two things together allow you to use DWYU in a reasonable fashion.

[arvid-norlander]

I should be able to find some time to test your new version next week, sorry for the delay.