Split internal extra calculation prec and BigDecimal.double_fig usage

[tompng]

BigDecimal's code was mixing up BigDecimal.double_fig(must be exactly 16) and internal extra calculation prec(could be any number around 10).
This change will clearly split this two essentially different constant usage.

Examples

This example should use double_fig because Float has double_fig precision. It must be 16.

initial_y = BigDecimal(Math.cbrt(x.to_f))
initial_prec = BigDecimal.double_fig * safe_factor
do_newton_method_iteration while prec < required_precision

This example should use EXTRA_PREC, a safe factor to reduce division/rounding error.
The value is completely unrelated to Float's internal precision. It can be reduced to 12, 10, 8 or something else in the future.

tan = sin(x, prec + EXTRA_PREC).div(cos(x, prec + EXTRA_PREC), prec)