Feature - Swaps parameters of sub reg, imm and removes neg

[Lichtso]

Currently the instruction sub reg, imm is useless because it can be encoded as add reg, -imm instead.
If we swap the parameters of sub reg, imm from reg = reg - imm to reg = imm - reg it becomes a useful instruction which would otherwise require two instructions to encode. Furthermore, the neg instruction can be removed as well because it can be encoded as sub reg, 0 (reg = 0 - reg) instead.

It seems that llvm is never emitting sub reg, imm for registers other than R11 (which was already addressed in #488).
For neg the instruction info in llvm would need to be patched here:
https://github.com/solana-labs/llvm-project/blob/7b8db05b564faffb522434b73b7082662171f94a/llvm/lib/Target/SBF/SBFInstrInfo.td#L319

[codecov-commenter]

Codecov Report

Merging #489 (6bc459c) into main (5debf64) will decrease coverage by 0.05%.
The diff coverage is 76.00%.

@@            Coverage Diff             @@
##             main     #489      +/-   ##
==========================================
- Coverage   89.69%   89.65%   -0.05%     
==========================================
  Files          23       23              
  Lines       10109    10126      +17     
==========================================
+ Hits         9067     9078      +11     
- Misses       1042     1048       +6     

Files Changed	Coverage Δ
src/ebpf.rs	79.59% <ø> (ø)
src/interpreter.rs	97.66% <66.66%> (-0.57%)	⬇️
src/jit.rs	92.63% <71.42%> (-0.27%)	⬇️
src/elf.rs	87.34% <100.00%> (+0.02%)	⬆️
src/verifier.rs	97.45% <100.00%> (ø)

[alessandrod]

I would use a more explicit swap_sub_operands() or something instead of
piggybacking on enable_neg. These flags are cheap and having an explicit one
makes the logic easier to follow.

And I would also add an if enable_neg() to the ::NEG_* handlers again for
explicitness, and so it doesn't have to rely on the verifier doing things at a distance