Fix unexpected Decimal sign behavior from Decimal(sign:exponent:significand:)

[EthanHumphrey]

Resolves #833. Fixes an issue that caused the initialized Decimal's sign not to match the passed sign when significand is negative. More information in the linked issue.

[xwu]

This would break clients who expect the documented behavior (including any Linux clients that have been able to use the fixed API for basically the whole time) rather than the buggy one. Importantly, there are synchronized changes that would also cause some clients expecting the buggy behavior to break.

For the purposes of preserving existing clients—important, I agree, given that my original bugfix took something like 4 years to be pulled into proprietary Foundation and originally pre-dated ABI stability—both this API and significand should have an ABI-stable entrypoint with the original behavior and the corrected behavior should be decorated with custom mangling.

However, I don't think a fix that changes the type to behave partly one way and partly another is it, as it is liable to create new and interesting incorrect behavior.

[pappalar]

Agreed the fix should not revert to the buggy behavior.
The actual fix is to make the RC follow ABI stability to not break existing clients with old toolchains

[EthanHumphrey]

Agreed with both points. I wrote this PR before discovering the original fix and the reasoning behind it. I'm in favor of fixing this to match the IEEE standard, but most concerned about older applications that might behave in weird ways without an update. The consequences could be very visible (in our app, all negatives turned positive), or could caused unexpected behavior in business logic not immediately clear to users. With iOS 18 available as of yesterday and it still not resolving this issue, I guess we will see quite soon what kind of impact it has.

Once a path forward is identified I'd love to update the PR for the experience.