non chunked algorithm using no regexps for faster run

[MagicalTux]

This is a re-creation of #2 with a separate source branch to avoid polluting the merge requests with stuff I've pushed to master on my fork.

I've modified this algorithm to be faster by not using regexps (too many Go projects tend to rely on regexp too quickly, losing a lot in performances) and also to limit allocations of heap memory as much as possible.

I've edited the README with a comparative benchmark, here is it:

Before:

goos: linux
goarch: amd64
pkg: github.com/MagicalTux/natsort
cpu: Intel(R) Core(TM) i7-8700K CPU @ 3.70GHz
BenchmarkSort1-12    	    6136	    300245 ns/op
PASS

After:

goos: linux
goarch: amd64
pkg: github.com/MagicalTux/natsort
cpu: Intel(R) Core(TM) i7-8700K CPU @ 3.70GHz
BenchmarkSort1-12    	  606818	      2013 ns/op
PASS

I've edited the tests to add one edge case (many zeroes prefixing a number), and all tests run OK, but I'd feel safer having more tests on this.