std: respect the bounds of the platform's data structures when performing time arithmetic

[joboet]

The docs for Instant and SystemTime mention that

An Instant is a wrapper around system-specific types and it may behave differently depending on the underlying operating system.

and

[checked_add returns] Some(t) where t is the time self + duration if t can be represented as SystemTime (which means it’s inside the bounds of the underlying data structure), None otherwise.

In my understanding, this implies that, should calling checked_add or checked_sub on an Instant/SystemTime succeed, it should always be possible to pass the resulting time value back to the operating system.

This expectation is currently violated on Windows and a number of other non-UNIX platforms (SGX, VEX, WASIp1, WASIp2, Xous) that immediately convert time values retrieved from the OS into Durations and use those Durations to perform arithmetic. Thus it becomes possible to create Instants/SystemTimes that are not representable by the OS, which becomes problematic when considering functions such as File::set_times or sleep_until.

This PR thus changes the Instant and SystemTime structures on the aforementioned platforms to store the data structure returned by their OS and changes the arithmetic functions to act on those values. In particular, this results

• in a loss of precision on VEX and Xous, as their clock only has micro-/millisecond precision (hence .checked_add(Duration::from_nanos(1)) might return the same time value)
• in a reduction of the representable range on all platforms (though the representable range is still larger than 100 years even on Windows, which is what is guaranteed by our documentation)
• in slightly changed performance characteristics on Windows, as the ticks of the performance counter will be converted only when performing arithmetic instead of when creating an Instant (this might actually be beneficial as Instant::elapsed will now only do one conversion instead of two)

I've left UEFI's Instant untouched in this PR, even though it is suffering from the same issue, as the fix is quite non-trivial there.

After this PR, the only platform where File::set_times can fail because of an invalid SystemTime is on 32-bit UNIXes that lack 64-bit time support. Changing them would conflict with the guarantee that 100 year additions are portable, however.

[rustbot]

r? @tgross35

rustbot has assigned @tgross35.
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