Change BigMath.sin and cos to always calculate in relative precision.

[tompng]

Almost all methods in BigDecimal/BigMath that takes precision calculates in relative precision.
Only BigMath.sin(except x around 0) and BigMath.cos(always) was calculating in fixed-point precision.
There is no document that says sin and cos is fixed-point precision. It can be considered bug (or wrong document).

BigMath.sin(BigDecimal('31.4159265358979323846264338327951111111111'), 30)
#=> 0.822691394060062514...e-31 (before, precision: 18)
#=> 0.822691394060062489417902505540769218359371379100137196517465784...e-31 (after)
#=> 0.82269139406006248941790250554076921835937137910013719651746578736517692e-31 (actual value)

Other

• Implementing BigMath.tan needs relative precision version of sin and cos.
• BigMath.tan(x, prec) will calculate in relative precision in all x around int * π / 2, so if sin and cos is in fixed-point precision, it's not consistent.