Fix CTS tests precision issues.

For sinpi, cospi, tanpi, expand the allowed range of errors by
doing a three step process:
1) compute a range of allowed argument*PI
2) for the minimum and maximum values of (1), compute the function
3) expand the resulting range by the allowed ulpFactor

For dot, dim, mad, mix, etc, we set a minimum error allowed.

For nextafter and powr, we now handle implementations that don't use
subnormal values.

Change-Id: Ife981f5bf31644d4726162ca5dd98e0c018fa7e3