Memoize should_trace for faster execution

[groutr]

This PR caches the output of should_trace which yields a slight performance increase.
Running the julia1_nopil.py benchmark on my laptop:
master: 8.14s
memoize_shouldtrace: 7.98s

Running the benchmark and looking a the cache statistics shows that the cache is hit 530 times for 1 file.

should_trace can perform two string searches before returning a result. os.path.abspath isn't exactly the speediest function either. Caching the results of should_trace makes a lot of sense.

I used the LRU cache available in the standard library. I feel the default maxsize of 128 is sufficient.

[emeryberger]

Good catch. Thanks!