Assess potential for compiler flag optimization

(Originally created in NCAR Teamwork 2015-11-11, but still relevant.)

We should assess the potential for speedups with more aggressive optimization, targeting particular modules that account for a substantial fraction of total run time. For example, could `-O3` or other optimizations help with CNZero?

I had been planning to see what we'd get by just turning on `-O3` for everything. But Jim Edwards convinced me that that is a Bad Idea: It sounds like there are actually relatively few places where both (a) -O3 has a significant benefit, and (b) using -O3 doesn't cause irreproducibility.

These optimization flags may be worth pursuing at some point, but it will be a process.