Speed up overlap checking (closes #131)

Ran on a medium-sized pytential FMM example (order 20, Laplace 2D kernel).

Before:

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
     8202    0.054    0.000  158.409    0.019 /Users/matt/src/loopy/loopy/symbolic.py:1806(do_access_ranges_overlap_conservative)

After:

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
     8202    0.030    0.000    4.771    0.001 /Users/matt/src/loopy/loopy/symbolic.py:1828(do_access_ranges_overlap_conservative)