Use rb_cstr2inum to improve performance

[evanphx]

This makes a big difference on Rubinius because it removes having to
create thousands of small Strings and the CAPI handles associated with
those small strings.

[chuckremes]

Please consider this patch. It vastly improves performance on Rubinius and looks like it improves MRI perf slightly too.

[evanphx]

I thought I'd include a benchmark that shows MRI is also quite a bit faster with this change:

» ruby scratch/cstr_bench.rb 
                user     system      total        real
parse:      6.550000   0.000000   6.550000 (  6.553141)
» gem install json-1.6.1.gem
» ruby scratch/cstr_bench.rb
                user     system      total        real
parse:      2.160000   0.000000   2.160000 (  2.165838)

given the benchmark: https://gist.github.com/1373994

[flori]

Thanks for this suggestion. I have used the fbuffer implementation from the generator in order to use the rb_cstr2inum method. I also used rb_cstr_to_dbl to parse JSON floats which is a bit faster as well. The benchmarking results are below (https://gist.github.com/1399957 for the floats):

Before:

(ffrank@office:json [master] ♢@json ->0)$ ruby -Iext:lib bench.rb 
                 user     system      total        real
parse:       4.260000   0.010000   4.270000 (  4.266780)
(ffrank@office:json ♢@json ->0)$ ruby -Iext:lib bench2.rb 
                 user     system      total        real
parse:      13.270000   0.030000  13.300000 ( 13.302467)

After:

(ffrank@office:json [master] ♢@json ->0)$ ruby -Iext:lib bench.rb 
                 user     system      total        real
parse:       2.550000   0.010000   2.560000 (  2.553988)
(ffrank@office:json [master] ♢@json ->0)$ ruby -Iext:lib bench2.rb 
                 user     system      total        real
parse:      10.530000   0.010000  10.540000 ( 10.546895)