[SR-6671] Fix Decimal implementation of FloatingPoint constants

[xwu]

This PR is a reprise of #1386, which has long ago gone stale. There are several issues with Decimal implementations of FloatingPoint constants:

First, as reported in SR-6671, the leastNormalMagnitude and the leastNonzeroMagnitude actually aren't. This is corrected in both the overlay and in swift-corelibs, with amendments to the corresponding tests.

Second, as pointed out by @stephentyrone in #1386, the implementation of leastFiniteMagnitude is incorrect. (It is also not a constant required by FloatingPoint, but with API stability it's too late now to remove from Decimal's API. It is left to the code owners as a separate matter whether they wish to deprecate this API.) Apple's documentation of the API states that it is "the decimal that contains the smallest possible non-infinite magnitude for the underlying representation." Magnitude is never negative, and the correct value here is zero. In this PR we make this API an explicit synonym for .zero. This is corrected in both the overlay and in swift-corelibs, with amendments to the corresponding tests.

[xwu]

@swift-ci test

[xwu]

@swift-ci test

[xwu]

@swift-ci test macOS

[xwu]

@swift-ci test

[xwu]

@swift-ci test

[stephentyrone]

(Just back from vacation, sorry for delay in reading this)

I'm worried about changing the value of leastFiniteMagnitude introducing subtle bugs to existing code that "works" with the current (incorrect) value. My advice would be instead to deprecate the property while keeping its value as is.

[xwu]

@stephentyrone Yeah, I as well. On the other hand, the value of the property is currently documented in a way that emphasizes its "magnitudeness" and there may be current code subtly broken relying on that. I think this should be deprecated regardless with a good error message, but it's not clear to me which value of the property is best in its deprecated state.

[stephentyrone]

Default assumption should be that already-compiled code is happy with the existing behavior (wrong though it may be), and we shouldn't break them. New code will get a deprecation warning, so we can catch them there. @spevans, how do you want to handle this?

(Note that leastFiniteMagnitude is not a value that comes from any protocol in stdlib or foundation, so there's little possibility of generic code misbehaving due to its nonsensical value.)

[xwu]

@stephentyrone See #3071.