Performance improvements

[telemenar]

This makes a number of performance improvements avoiding largely copying and additional comparison.

The two most significant pieces are:

• re-arranging to avoid needing to allocate memory for splitslash
• switching to an enum for the mode

[ptzz]

@telemenar This is good stuff, thanks! I forked your repo to https://github.com/ptzz/ctrlp-cmatcher and optimised further. Managed to cut the processing time by ~80% for large input. The delay is not really noticeable now for 100k files on a 2012 Macbook.

The trick was to use partial sort and only sort the items visible on screen.