Improve the build performance of mp_find.

[Mizuchi]

This improves the build time by looking up 100 elements in a batch.

Note this is just an optimization. It won't change any observable behavior. In addition, this is disabled for GCC due to a gcc bug.

Tested by creating a list with 996 elements and call mp_find: https://gist.github.com/Mizuchi/fce64eaa400288de136be1181f0edc96

The mp_find_2.cpp build time reduced from 400ms to 80ms.

It's possible to re-implement this with boost.preprocessor to make the batch size (100) configurable, but that will introduce boost.preprocessor as a dependency.

[pdimov]

I don't think that's going to work if the list contains duplicates.

[Mizuchi]

I don't think that's going to work if the list contains duplicates.

Good point. It's fixable by (ab)using partial specialization order (I created a new commit for demo).

This can still reduce the build-time from 400ms to 140ms (for the mp_find_2.cpp example), though that makes the implementation much more complex.

Let me think about whether there is a better way to do. I'll close this pull request for now.