Fix FScale'ing zero that should return zero

[pmatos]

Added tests will fail without the current patch.

[pmatos]

This pcem issue sarah-walker-pcem/pcem@2e037b5#diff-56c35b563dbfc08142a6ac0ee8bd3c626b5556065b41f39732fdb3625982064cR677 made me look into this in FEX.

The table in https://www.felixcloutier.com/x86/fscale clearly shows that fscale should return zero unless st0 is either NaN or +inf.

[Sonicadvance1]

Nice catch!

[Sonicadvance1]

Had a thought, does this return the correct sign of zero?

[pmatos]

Had a thought, does this return the correct sign of zero?

Good point! It doesn't. I assumed it didn't matter. I am still unsure how it deals with the exceptional cases of st1 being +inf or Nan, which return NaN, or when st1 is -inf, in which case it returns -/+ 0. There's an odd asymmetry here. Unfortunately, I cannot run MDK2 yet.

[pmatos]

Had a thought, does this return the correct sign of zero?

It does now! :)
Updated code and tests.

[pmatos]

Had a thought, does this return the correct sign of zero?

It does now! :) Updated code and tests.

ugh - clearly something is not working 100%.

[pmatos]

I am slightly dumbfounded that after much searching, the crashes were due to an include of softfloat_types.h into SoftFloat.h. To make matters worse, my editor keeps wanting to add it to the changeset without letting me know. Really need to understand which feature this is and disable it.