Varint optimizations

[zanker]

Per #268, here's the results of benchmarking C vs pure Ruby. For the sake of being thorough, I also tested the variant used in the other protobuf gem (which was related to what @jjowdy) suggested, and it performed worst.

       user     system      total        real
pure_protobuf  1.080000   0.010000   1.090000 (  1.087074)
pure_alt  1.260000   0.010000   1.270000 (  1.275954)
c  0.700000   0.000000   0.700000 (  0.708581)

I also tested the C encoder in the varint gem, and it appears to be slower. This might be because with how protobuf is implemented, we need to create a StringIO class, rewind it, and then call read in order to get. Rewriting things might speed it up, but that's outside of what I'm looking to do in this PR.

Thanks @hollow and @codekitchen for the extension!

[zanker]

I picked pretty arbitrary values to test that these are working. If someone has another suggestion, happy to include something else.