Increase performance of browscap 3.x

[asgrim]

As per conversation with @jaydiablo here, I'd like to review the performance of the 3.x series.

Although we have significantly reduced the memory usage in 3.x, it now takes much longer to process.

@jaydiablo has pinpointed a potential big win:

This was the biggest win for me when I was trying to optimize crossjoin: https://github.com/crossjoin/Browscap/commit/2ee9663bd4b971c658d87f33fe780858580f7bd4 (the digit replacement in that commit is a bit overzealous, and causes tests to fail, which was fixed later https://github.com/crossjoin/Browscap/commit/afd9f8eaec3fe506ed864a93a536e2370b570922)

Investigate this and see if we can optimise similarly.

[mimmi20]

The changes from crossjoin/Browscap@2ee9663 were reimplemented in the optimize branch and merged at the 26.09.