[llvm] [DAG] Introduce generic shl_add node [NFC] (PR #88791)

[Simon Pilgrim via llvm-commits]

https://github.com/RKSimon commented:

No objections from me - needing MUL_IMM at all has always annoyed me, its often needed to stop extra combines that would prevent us matching for LEA etc. but that often leads to more complex code than necessary.

There's the possibility we could use SHL_ADD for vector gather/scatter address math (vector index + offset, imm scalar shift amt), but I don't see anything here that would prevent that in the future?

Note - there are a few real clang-format warnings that need cleaning up.

https://github.com/llvm/llvm-project/pull/88791