Introduce O(n^1.5) BigDecimal parser implementation

[ferenc-csaky]

Covers the BigDecimal part of #577. The double/float implementations would require a much bigger refactor, since the example implementation follows a very different approach, but IMO it's worth to update only the BigDecimal without the others as well.

[cowtowncoder]

@ferenc-csaky Sounds good -- thank you very much for the contribution! -- and it looks like I have CLA from you already.

Just one request: could this be rebased against 2.13? master is for Jackson 3.0 and I think this improvement would be great for 2.13 -- I can handle merging to master.

[ferenc-csaky]

Changed the base branch to 2.13. My original plan was this as well, my dev. branch also originated from 2.13, but it seems I missed it in the PR. Corrected now, thanks!

[cowtowncoder]

Just realized that none of the tests verifies new code path(s), I think. Will repurpose some from original... should have caught this earlier,.

[cowtowncoder]

@ferenc-csaky Hmmh. I think there may be a bug somewhere, some offset not being carried or used for calculations?
I added a test in 2.13 from the original changes and it looks like offset used for splits is 0 for some reason.
As far as I can see it gets passed correctly to BigDecimalParser (first suspected it might not but seems to).
Seems to only happen with character-backed parser, not byte-based (probably former will always have 0-based buffer for text value).

One thing that is tricky, too, is how to propagate underlying secondary fails, if those are possible (not sure if they are after fixing this).

[cowtowncoder]

Looks like int numIdx = 0 in parseBigDecimal(int splitLen) is wrong: that should get set to off, somewhere. But simple attempts did not produce expected results.

[ferenc-csaky]

Oh yes, you are totally right, I didn't follow the source impl. exactly and I messed it up. Sorry about it. I just created a PR with the fix. Now the test you added runs correctly.

[cowtowncoder]

@ferenc-csaky np. Just glad this was caught :)