Implement Liquid::ResourceLimits

[peterzhu2118]

Implements Liquid::ResourceLimits in C. It is roughly a translation of the Ruby code into C. I was not able to measure a significant speedup on the benchmarks.

[dylanahsmith]

I think the performance improvement would mostly be significant within a capture tag at the moment. So you could try temporarily modifying the benchmark to wrap the benchmarked templates in a capture tag.

I expect it will also help with the assign tag and to reduce the instance variable lookup overhead on the resource limits object.

[dylanahsmith]

We can do any initialization here that doesn't depend on the initialize arguments and doesn't need to allocate any objects. In this case, we could even leverage the reset function for that purpose.

[dylanahsmith]

I still think we should move resource_limits_reset(resource_limits); from the initialize method to this allocate function, which will also remove the need for having to use ResourceLimits_Get_Struct(self, resource_limits); there.

[peterzhu2118]

Thanks for the feedback @dylanahsmith! I changed the benchmark files to use capture tags, I think there is a slight improvement in performance, however the results are pretty noisy on my computer.

Master:

              parse:    129.494  (±10.0%) i/s -      1.287k in  10.046882s
             render:    135.550  (± 8.9%) i/s -      1.356k in  10.093187s
     parse & render:     62.487  (± 6.4%) i/s -    624.000  in  10.033199s

This branch:

              parse:    131.418  (± 6.8%) i/s -      1.313k in  10.041110s
             render:    149.923  (± 8.7%) i/s -      1.498k in  10.075368s
     parse & render:     65.270  (± 6.1%) i/s -    654.000  in  10.067960s

[dylanahsmith]

Wow, that is a more significant difference in the benchmark than I expected.

[dylanahsmith]

I still think we should move resource_limits_reset(resource_limits); from the initialize method to this allocate function, which will also remove the need for having to use ResourceLimits_Get_Struct(self, resource_limits); there.

[dylanahsmith]

A couple of minor things, otherwise, LGTM

[dylanahsmith]

In #64 (comment) I meant that this should be done in resource_limits_increment_write_score_method to avoid doing it redundantly in the VM loop.

[dylanahsmith]

This is unused, so can be removed now.