Add faster substring search

[constinit]

This commit implements the fast substring search algorithm described by
Leonid Volnitsky.

The full algorithm is described at:
http://volnitsky.com/project/str_search/

Briefly, we insert each consecutive pair of characters in the needle
into a hash table. We then step through pairs of characters in the
haystack. If the pair is in the hash table (and probably also the needle)
we do a character-by-character comparison.

Performance is 30-40% faster than Boyer-Moore on 2GB of text:

./ag_hash --stats blahblahbla ~/Downloads/wiki/
0 matches
0 files contained matches
4 files searched
1956034523 bytes searched
0.800452 seconds

./ag_master --stats blahblahbla ~/Downloads/wiki/
0 matches
0 files contained matches
4 files searched
1956034523 bytes searched
1.310972 seconds

[ggreer]

This is a pretty cool idea, but it segfaults on my system when I try to search my code directory. Have you tried it on any large corpus of data?

[constinit]

Thanks for reviewing. I think I solved the problem — I was overrunning the file buffer because I thought it's null terminated.

I tried a randomized testing approach on a large corpus, and I haven't managed to crash it since (except for unrelated issue #741). I encourage you to take another look.

[ggreer]

Thanks a lot for this PR. You've improved the speed of ag more than I've managed to in the past two years!