Testing is Sensitive to Optimization Flags

[mandli]

@rjleveque noticed that tests were not passing for him due to assertion failures. I did a quick cross-section of different optimization flags to see how wide-spread the problem was (see table below).

Optimization flags up to -O1 are fine. The compiler gcc defaults to -O0, which confusingly does not turn off all optimizations (see here for more info). OpenMP does not seem to have anything to do with the tests not passing thankfully. There is an interesting thing I did notice with the recursive rpn problem we have seen before. This comes up and a make new fixes things, but this seemed to occur when changing optimization levels. May be a clue here as to why that occurs, although it is an easy fix.

I propose that we explicitly list the compiler flags in the testing so that what is being tested is clear. We can also have a matrix of testing so that we can test out different flag combinations or compilers if we really wanted to.

Working	Flags	Failed Tests	Notes
✅	None (default)		FFLAGS=-O0 we think
✅	Debug = -O0 -W -Wall -fbounds-check -fcheck=all -pedantic-errors -Wunderflow -fbacktrace -ffpe-trap=invalid,zero,overflow -g	test_acoustics_2d_adjoint_forward, test_acoustics_2d_adjoint	This is due to a bug #304, with it incorporated everything passes
✅	OpenMP = -fopenmp, OMP_NUM_THREADS=8	test_advection_2d_annulus, test_advection_2d_inflow, test_advection_2d_square	Recursive problem, fixed by recompile
✅	hybrid = -fopenmp -O, OMP_NUM_THREADS=8	test_advection_2d_annulus, test_advection_2d_inflow, test_advection_2d_square	Recursive problem, fixed by recompile
✅	hybrid = -fopenmp -O, OMP_NUM_THREADS=1
✅	FFLAGS = -O1 -fopenmp, OMP_NUM_THREADS=10		Had to fix the recursive problem again, maybe something to do with optimization flags?
	FFLAGS= -O2 -fopenmp, OMP_NUM_THREADS=1	test_acoustics_2d_adjoint_forward, test_acoustics_2d_radial	Adjoint problem is off by comparison and radial has two more times recorded
	FFLAGS = -O3 -funroll-loops -finline-functions -march=native, OMP_NUM_THREADS=10	test_acoustics_2d_adjoint_forward, test_acoustics_2d_radial	Adjoint problem is off by comparison and radial has two more times recorded