Manually confirm that optimization settings do not have significant effects on results

[axch]

If I change output_preservation_test.py to compile the MIM core with -O1 or -O2 rather than -Ofast, two undesirable things happen:

• All three of the n-layer simulations crash with a memory corruption error (*** Error in 'MIM': double free or corruption (!prev): 0x00000000007c8a90 ***).
• Two of the reduced gravity simulations give outputs that disagree with the known good ones.

The former of these is Issue #28, and is certainly more urgent; this ticket is about confirming that the disagreements are small, and caused by liberties gfortran -Ofast takes with the IEEE 64-bit floating point standard. Presumably the mechanism for doing that would be implementing the first option outlined in Issue #27.

[axch]

PR #35 permits one to easily manually check whether a different compiler option produces results that are too far from the known-good ones.

Down-scoped this issue to the manual part so it can be closed, and filed full-scope automatic platform-induced discrepancy testing as #36.