Minor float/float32/double/float64 cleanup

[derobins]

• Remove noise casts
• Use FLT_EPSILON and DBL_EPSILON instead of misc epsilon guesses
• Add F suffix for float constants