Request to cherry-pick a57a7f3de551 onto release/14.x

[david-arm]

This patch fixes a crash in AArch64TargetLowering::ReconstructShuffle when we encounter a scalable vector, and just bails out of the optimisation if this happens. The patch only affects scalable vectors, and we are doing our own additional release testing with scalable vectors for LLVM 14.x.

[david-arm]

/cherry-pick a57a7f3

[llvmbot]

@llvm/issue-subscribers-backend-aarch64

[llvmbot]

/branch llvmbot/llvm-project/issue53716

[llvmbot]

/pull-request llvmbot#31

[tstellar]

@TNorthover What do you think about backporting this? a57a7f3

[TNorthover]

Yep, it's another clear tightening of where SVE is allowed.