Allow branch instructions inside loops

[dop-amin]

Still to do:

1. Automatically adding the branch tag. I was not sure where the right place for this is so I skipped that for now. I feel like I could just "throw" it into a lot of places but it always felt oddly specific. Perhaps in the constructor of the Loop subclass, as this is most closely related.
2. Fixing the selftest. Not 100% sure what we'd like to have here. We could in theory also run the loop in the test, I suppose but you have a better overview of how this all works at the moment. Filtering out the branches before calling the selftest is again possible, yet again, kind of a "specific" thing.
3. More proper testing. I ran a few examples but there are many I haven't tried so far.

[mkannwischer]

Still to do:

1. Automatically adding the branch tag. I was not sure where the right place for this is so I skipped that for now. I feel like I could just "throw" it into a lot of places but it always felt oddly specific. Perhaps in the constructor of the Loop subclass, as this is most closely related.
2. Fixing the selftest. Not 100% sure what we'd like to have here. We could in theory also run the loop in the test, I suppose but you have a better overview of how this all works at the moment. Filtering out the branches before calling the selftest is again possible, yet again, kind of a "specific" thing.
3. More proper testing. I ran a few examples but there are many I haven't tried so far.

I have implemented (1) and (2) - please have a look.
(3) is still a todo. In particular, we need to re-optimize all examples and make sure their performance does not degrade. I did make sure everything runs successfully, but I do see some performance degradation (and some performance improvment) and some breaking tests. This needs further investigation.

[dop-amin]

Thanks, your changes look fine to me!

On the performance: I ran the optimizations on main as well as on this branch multiple times and observed "the usual" deviations, sometimes one result was better, sometimes the other. I'd deem this okay; at least I'm assuming no degradation due to this PR.

On the broken tests: I was able to reproduce this, either wrong result data being printed in the end or seemingly endless garbage that gets send over UART. There have been two issues: 1. In small_pointmul_asm_769, the register order in cmp was not adhering to what we require. This caused the loop counter fixup to be wrong which in return made the loop have one more iteration than it should causing in OOB write. 2. In basemul_257_dilithium the ldr with increment of the loop "counter" got moved after the cmp, meaning we also had an extra iteration here. Reason is the address offset fixup, removing the outputs from the ldr. Fixed this by disabling the fixup. This can potentially also happen in other examples.

I address the two things in 8674344. If you think this should be part of this PR, good, if not, feel free to drop it and we handle it separately.

[mkannwischer]

Please re-optimize the examples we already have in main.
Rest looks good to me! Thank you!

[dop-amin]

Re-optimized code is added, tests on this code pass and no more weird output.