Allow instruction combine in SBF

[LucasSte]

This PR removes code that added a passthrough function in SBF (they were originally from the BPF target), which prevented LLVM's instruction combine pass to run. Their original purpose was to appease the kernel verifier. As a result, we now combine jumps.

Old code generation:

lsh64 r1, 32
arsh64 r1, 32
mov64 r2, 1
jsgt r2, r1, LBB0_3
jsgt r1, 7, LBB0_3

New code generation:

add64 r2, -8
lsh64 r2, 32
rsh64 r2, 32
lddw r3, 4294967289
jgt r3, r2, LBB0_2