Use more fortran intrinsics

[Copper280z]

This changes a bunch of loop based array initializations to use fortran array assignments, which has 2 benefits, less clutter in the codebase, and the compiler better optimizes these. This also changes several calls of MATMULT_FFF and the transpose version to use the fortran intrinsics MATMUL and TRANSPOSE. This change is worth roughly a 10 second improvement on the big plate benchmark problem on the two machines I tested on.

There's a lot more places that use the loop based initialization. Same with MATMULT_FFF, I purposefully kept that narrow on areas I could test because there are places where you need to think a bit to keep the behavior the same, ie if the arrays aren't trivially broadcastable.

Edit: This should go in a separate PR:
The other main change is to disable the windows-XP era array size limit (~2GB) by setting it to 100TB. For some users/problems this may expose a preexisting issue that was masked by this limit: mystran overallocates the stiffness matrix to be the maximum possible size based on the number of elements times the number of elements in their dense stiffness matrices, then after construction it frees the unused memory. It was pretty easy to hit this limit, so mystran would allocate less than "intended" but end up with a global stiffness matrix much smaller than even the artificially low limit. Now mystran will allocate the full amount based on the conservative estimate. This is only a practical concern now due to the recent performance improvements, so it seems unlikely to impact many end users. The fix here is either a better bounding estimate or some sort of reallocation strategy with a less conservative estimate.

[Copper280z]

Restructured the commits into initialization and matmul/sliced assignment, dropped the IF reordering and winxp memory limit change, those should be separate from this PR. Also added a few more matmul and initialization updates in the files already modified.

I also double checked assignments on allocatables, I believe all now explicitly slice the array to the same size as the previous loop method, this includes any matmuls that were implicitly sliced before.