[llvm] [LV] Add a flag to conservatively choose a larger vector factor when maximizing bandwidth (PR #156012)

[Yuta Mukai via llvm-commits]

ytmukai wrote:

This is a patch to resolve one of the performance regressions that occur when maximizing vector bandwidth for AArch64 SVE. (https://github.com/llvm/llvm-project/pull/109671#issuecomment-3077403635)

https://github.com/llvm/llvm-project/pull/156012