[llvm] [LV][SVE] Recognize potential DOT sequences and use a wider VF (PR #69587)

[Graham Hunter via llvm-commits]

huntergr-arm wrote:

This patch sets up for a target-specific pass (https://github.com/llvm/llvm-project/pull/69583) to generate SVE SDOT/UDOT instructions when appropriate.

This is some downstream code I wrote 4 years ago that we want to upstream -- suggestions on alternative approaches welcome.

https://github.com/llvm/llvm-project/pull/69587