Implement round-ties-to-even for Duration Debug for consistency with f64

[josh-kaplan]

Summary

This PR proposes a fix for #103747 implementing IEEE-754 S4.3 roundTiesToEven for Duration Debug implementation.

Testing

Added new test in time.rs for validating roundTiesToEven behavior in Duration formatting. Reran all debug formatting tests in time.rs with ./x test library/coretests --test-args time::debug_formatting.

[rustbot]

r? @scottmcm

rustbot has assigned @scottmcm.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[hkBst]

Maybe also add a test-case for println!("{:.0?}", Duration::MAX);, unless it is already there?

[josh-kaplan]

Maybe also add a test-case for println!("{:.0?}", Duration::MAX);, unless it is already there?

Fair! It looks like it already exists on line 330 in debug_formatting_extreme_values.

I also updated the comment above the new rounding behavior to accurately reflect the behavior change.

[scottmcm]

r? libs

[Mark-Simulacrum]

Hm, I hadn't realized how complex the logic already is for Duration's Debug impl. I'm not convinced that there's a strong use case for being this careful in a Debug impl for Duration -- no one should be parsing those outputs, IMO, or relying on particular rounding there. This differs from Display/Debug for floats which are much more likely to be in the critical path for correctness.

That said, the diff here doesn't seem too complicated, so maybe it's OK, especially if we can simplify it a bit (per my inline comment).

View changes since this review

[Mark-Simulacrum]

It would be nice to merge this code with the below - in particular to not duplicate or miss having the useful comments. It seems to me that if we modified is_last_digit_odd in the if here to be or'd with fractional_part > divisor * 5 that would work out? Or am I missing some case?