atan2(2,0.3) gives non-float answer /FIX

atan2(a,b) should be an approximate number (float or
bfloat) whenever one of the arguments is approximate
and the other is either approximate or an exact number.

This is not currently true. All the following give results
involving %PI: atan2(-1.0,-1), atan2(-1.0b0,-1.0), atan2(-
2/3,-1.0b0) etc.

The attached code fixes this.