More BCMath performance improvements

[ndossche]

Builds upon all previous PRs.

With benchmark from #14076.
Before:

1.5271680355072
1.9233000278473
2.0597560405731

After:

1.4276769161224
1.8403890132904
1.9751319885254

One of the changes here is avoiding initializing and releasing temporary numbers, but only done for the most simple (and common) functions.

[SakiTakamachi]

I made one comment, but LGTM

[SakiTakamachi]

Suggested change

	result = bc_add (first, second, scale);
	bc_num result = bc_add (first, second, scale);

Maybe can declare and initialize it at the same time and remove the result from the cleanup.

But I'm not sure if it affects performance.

[ndossche]

Good question. I expect this will only make the failure path potentially faster, as it can avoid a call then to the freeing. For the success path I don't think it matters, but I guess feel free to try.

[Girgias]

Changes make sense to me.

Please rebase to keep commit history