Address performance issue with `BigDecimalParser`

[cowtowncoder]

Although there were reasons to expect BigDecimalParser could help handling of very large numbers -- and was consequently added for 2.14 via #814 -- there are concerns that there are cases where its performance is sub-standard.

In particular there are concerns with use of large exponents (scientific-notation) which lead to "big" numbers but ones that are compact enough (wrt textual representation) not to trigger "too big number" guardrails but lead to poor performance.
If and when regular JDK BigDecimal parser does not have similar issues, we may need to remove usage.

A test case exists to show the issue but I will not include it here: contact me over mailing list(s) for details if interested.

[cowtowncoder]

/cc @pjfanning This is one of 2 that we had discussion about

[pjfanning]

Might be useful to backport this to 2.14.

[cowtowncoder]

@pjfanning I guess if it's safe enough, my only concern would be about breaking something. Then again, functionality itself is probably rarely actually used in real-world production (fwtw :) ).