Support Call::dynamic_shuffle for LUT in SVE2

[stevesuzuki-arm]

OptimizeShuffles pass is enabled in CodeGen_ARM for SVE2.

• Detects gather load where index range is bounded within certain value e.g. Look Up Table
• Transforms it into contiguous load + Call::dynamic_shuffle, which is lowered to TBL instruction by codegen.

This is especially useful to vectorize with long vector in SME2 streaming mode where general form of gather load is unsupported.

OptimizeShuffles is modified so that we can use it commonly between targets (for now, Hexagon and ARM SVE2).

Checklist

• Tests added or updated (not required for docs, CI config, or typo fixes)
• Documentation updated (if public API changed)
• Python bindings updated (if public API changed)
• Benchmarks are included here if the change is intended to affect performance.
• Commits include AI attribution where applicable (see Code of Conduct)

[abadams]

Does this happen only for allocations we've made ourselves, or also for input buffers? I see that if pads our own allocations but I don't see that it skips this for input buffers. I think hexagon might play fast and loose when it comes to reading out of bounds on input buffers by up to a vector. I fear in general this will fault if it crosses a page boundary though.