Stop using logarithm to compute the number of components.
Instead, use the theoretical maximum number of components for buffer,
and count up the actual number of components during conversion.

https://github.com/ruby/bigdecimal/commit/9067b353ac