Improve the performance of s-shared-start

[cpitclaudel]

This yields a 2-3x speedup on short strings and 5-15x speedup on longer ones on
my machine. I haven't done in-the-wild benchmarking because I don't currently use s.el, but I noticed this while reading the implementation. Here's a mico-benchmark:

(defvar ~/s1 "/home/clement/.emacs.d/.cask/28.0/bootstrap/s-20180406.808/s.el")
(defvar ~/s2 "/home/clement/.emacs.d/.cask/28.0/bootstrap/commander-20140120.1852/")
(benchmark-run-compiled 100000 (old-s-shared-start ~/s1 ~/s2))
;; Right after GC: (1.114678701 0 0.0); Normal use: (1.2542185909999999 1 0.13476934700000243)
(benchmark-run-compiled 100000 (new-s-shared-start ~/s1 ~/s2))
;; Right after GC: (0.072596075 0 0.0); Normal use: (0.22450463 1 0.1343407259999907) 

A similar trick could be used for s-shared-end, but I don't know of a way to do it without reversing the input strings (it's still ~2x faster, but it allocates, so the benefit is less clear-cut).

Do you need an update to the contributors list as well for the PR?
Also, I have copyright papers for Emacs on file, if that's needed.

[alphapapa]

This looks like a nice patch. :)

[cpitclaudel]

Ping on this.

[magnars]

Thanks!

[magnars]

And thanks for the ping. :)