Investigate functools.lru_cache #1125

I'm not sure if functools.lru_cache existed when we added galsim.utilities.LRU_Cache, but I've been playing around with it recently and I think I like it better. With this simple program:

from functools import lru_cache
from galsim.utilities import LRU_Cache
import numpy as np

@lru_cache(1024)
# @LRU_Cache
def f(x):
    return x*2+np.sin(3*x)+np.cos(5.5*x+0.2*x)**2

def g(x):
    return f(x) + f((x+31)%1000)

def main():
    for i in range(1000000):
        g(i%1000)

if __name__ == '__main__':
    main()

I get better performance with functools.lru_cache (~1.0 s vs ~1.8 s), and also like the output I get through gprof2dot better:

utilities.LRU_Cache

functools.lru_cache

Querying the cache hits/misses is also nice with functools.lru_cache.

The one feature we'd be giving up is the ability to resize the already-created cache. It's definitely nice for the user to be able to set the size, so I think we'd need to invent some kind of API wrapper for that. Looks like there's a __wrapped__ attribute that might help.

Finally, functools.lru_cache is only available for python >3.2, so this would be easiest if we dropped support for 2.7.