Cortex-M7 Kyber Part 2: NTT functions

[dop-amin]

This PR adds Kyber's NTT functions. The forward NTT switched to the Arch_Armv7M.BranchLoop type in order to get rid of the spilling feature, which allowed the vmov+cmp loop without boundaryReserverdRegs.

Possible improvments:

• Make address fixup work with the current code.
  • Problem 1: Split heuristic cannot handle an instruction not being found by id when enforcing a before constraint. Could likely be resolved by allowing the constraint to be optional such that in case the instruction with a certain id can't be found AND we're using splitting heuristic, we continue instead of failiing.
  • Problem 2: Sometimes, instructions still get reordered after the branch which completely breaks the code. Needs investigation.
• Run with more liberal parameters.
• Also use Arch_Armv7M.BranchLoop in the inverse NTT.

[dop-amin]

The first point in the list should now be addressed by #150 and #149.
I re-optimized the code using those changes and (partially) enabling the fixup while switching both examples to the Branch loop type.

This PR now depends on #150 and #149.

Tests in pqmx passed and the performance is similar (+/- 50 cycles roughly) to the armv7m branch.