[X86] Mark some floating point operations that are always expanded for vector...

[X86] Mark some floating point operations that are always expanded for vector types as Expand in a floating point only loop instead of looping through all vector types.

llvm-svn: 266850