Compiled version of DelayedCall

[itamarst]

I ran an experiment where I made a Cython version of DelayedCall, with reset() being compiled method. The result was a slightly faster HTTP server benchmark (it creates a timeout per request, with usage of reset()). This suggests there is room for performance improvements here.

I suspect there's rather more one could do, e.g. the base reactor has a sorted list of DelayedCall which relies on doing Python based __gt__/__lt__ comparisons between objects, and... that could all be done internally with a faster data structure that doesn't call into Python for comparisons. But that may be a larger project.