Vectorize acos, asin, and atan for float.

Reference issue

What does this implement/fix?

This change vectorizes the acos, asin, and atan operators in Eigen.

Additional information

Accuracy: Exhaustive testing for all float arguments in [-1:1] shows that this implementation is accurate to 2.6 ulps for pacos, and 3.8 ulps for pasin. Maximum relative error for patan is 2 ulps.

Speed: See: $2393114

Speedup for 4k element vector:

	SSE	AVX	AVX512
asin()	6x	15.9x	18.6x
acos()	11.4x	29.5x	29.8x
atan()	5.5x	9.3x	17x