[X86] Split WriteFSqrt+WriteFDiv into WriteFSqrt64/WriteFSqrt32 + WriteFDiv64/WriteFDiv32

[RKSimon]

Bugzilla Link	36909
Resolution	FIXED
Resolved on	May 23, 2018 08:05
Version	trunk
OS	Windows NT
Blocks	#36256
Fixed by commit(s)	331629,331643

Extended Description

It's highly unlikely that a target can perform single and double sqrt/div at the same rate, so why do we put them under the same scheduler class?

x87's fdiv instructions should probably be put under WriteFDiv64 and customized if necessary.

[RKSimon]

WriteFSqrt was split into f32/f64/f80 variants at rL331643

WriteFDiv was split into f32/f64 variants at rL331643