[X86] Poor codegen for NE/EQ zero/allones comparisons for very large integers

[RKSimon]

Building on #163373

For integers larger than the maximum legal vector width, we should be able to split into smaller vectors and then AND/OR them together before performing the comparison

[llvmbot]

@llvm/issue-subscribers-backend-x86

Author: Simon Pilgrim (RKSimon)

Building on #163373

For integers larger than the maximum legal vector width, we should be able to split into smaller vectors and then AND/OR them together before performing the comparison