Use widen(Int128)=BitIntgers.Int256 within FD

[NHDaly]

Remove allocations for FD{Int128} by switching to widen to Int256 instead of BigInt.

Adds a dependency on the BitIntegers package, but it seems worth it to me.

I'm merging this into #5, so that we can review them in two chunks.

[NHDaly]

Here are the benchmark results from this change -- the improvements are really big! :D

(The diff is relative to JuliaMath@1768c58.)

		Operation	Values
		*		/		+		div		identity
Category	Type	time (ns) *	% improvement	time (ns) *	% improvement	time (ns) *	% improvement	time (ns) *	% improvement	time (ns) *	% improvement
Int	Int32	0.35	0%	3.91	-3%	0.35	0%	0.35	0%	0.35	0%
	Int64	0.35	0%	3.91	-4%	0.35	0%	0.35	0%	0.35	0%
	Int128	0.35	19%	13.45	-4%	0.35	0%	0.35	0%	0.35	0%
Float	Float32	0.35	0%	0.35	0%	0.35	0%	2.57	4%	0.35	0%
	Float64	0.35	0%	0.35	0%	0.35	0%	2.67	-4%	0.35	0%
FixedDecimal	FD{ Int32,2}	3.02	71%	10.25	-18%	0.35	0%	0.35	0%	0.35	0%
	FD{ Int64,2}	9.54	47%	22.88	-9%	0.35	0%	0.35	88%	0.35	0%
	FD{Int128,2}	73.46	99%	1366.20	90%	0.35	100%	0.35	100%	0.35	100%
Big	BigFloat	93.60	-32%	167.97	-20%	52.76	-25%	193.86	-26%	0.35	0%
	BigInt	186.73	8%	553.34	-14%	174.12	5%	229.65	-15%	0.35	0%

* note: The time is truncated to a minimum of 0.345ns, which is the estimated time for a single clock cycle on my machine

[tveldhui]

Is this used? If not consider removing?

[NHDaly]

Ah yes it is. I've added a comment about why it's used. (_round_to_even calls iseven, which in turn calls isodd.)

[tveldhui]

What is XBS? eXtra Big Stick?

[NHDaly]

Haha i guess it's "eXtended Bit (integers) Signed". Blech!

But also, i just switched it to XBI, which is "eXtended Bit Integers", which is what I settled on in the PR i sent to that package: rfourquet/BitIntegers.jl#2

It's basically the super type for all the new integer types that package added.

I clarified that in the comments, and also added a TODO comment to delete these lines once the PRs on the base package are merged. Thanks! :)

[NHDaly]

(Just force-pushed after rebasing in the comments on #5)

[NHDaly]

(Back to you!) Thanks! :)

[tveldhui]

comment: force evaluation of max_exp10 at compile time ?