[llvm] [SDAG] Heed enableAggressiveFMAFusion when folding fma(a,b,c*d)+e (PR #94209)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> (assuming fadd/fmul/fma all have the same performance)

This should be controlled by isFMAFasterThanFMulAndFAdd? Not exactly sure what the parallel control enableAggressiveFMAFusion means. It's even faster? 

https://github.com/llvm/llvm-project/pull/94209