Adaptive interval

[joerick]

Idea: an 'adaptive' interval, that pyinstrument could lower the interval as the profiler session progresses, such that you always have e.g. 0.1% accuracy. e.g. start each session at 1ms accuracy, then when you get to 10 second duration, switch to 10ms, then at 100 seconds, switch to 100ms, etc.

Would fix #169, I think.