[llvm] [RISCV][GlobalISel] Call `tryEmitMemcpyInline` for G_MEMCPY_INLINE (PR #192671)

[Matt Arsenault via llvm-commits]

================
@@ -154,6 +155,19 @@ bool RISCVPreLegalizerCombiner::runOnMachineFunction(MachineFunction &MF) {
   return Impl.combineMachineInstrs();
 }
 
+bool RISCVPreLegalizerCombinerImpl::tryCombineAll(MachineInstr &MI) const {
----------------
arsenm wrote:

I don't remember exactly how it ended up like this. I think it ended up as an optimization for memcpy, and then repurposed for the mandatory inline lowering. I'm not sure there's any reason to maintain the combiner expansion paths 

https://github.com/llvm/llvm-project/pull/192671