Fix sequence generation to be always increasing

[Qqwy]

• Fixes generation of the normal generator
• Fixes generation of the sync generator
• Replaces the two incorrect test_generate tests with test_generate_ordered tests
• Gets rid of MaybeUninit inside the sync generation. I've looked at the created assembly (using cargo asm) and the compiler is more than clever enough to optimize the wrapping and unwrapping into an Option away here, so there's no need for unsafe. (Extra fun as it was the only unsafe in this crate, so now it's fully no-unsafe!)

Fixes #5

[MrGunflame]

Thanks for the PR, please run cargo fmt on the thing once. Looks almost fine besides that.

[Qqwy]

Thanks for reviewing 😊 ! Your feedback has been applied.

I'm running the loom tests locally now (I did not realize that they didn't run as part of the normal test suite before), using RUSTFLAGS="--cfg loom" cargo test --lib --all-features --release, and see that the no_duplicates_no_wrap test is failing.
I'll investigate what is going on there.

[Qqwy]

The trick was that since the sequence stores the 'next' value, at the very start we have to initialize it at 1. Otherwise, in the edge case where someone tries to generate a snowflake in the very same millisecond as the epoch, it would immediately trigger the wait condition.

In real programs this won't ever come up of course, since time will always have passed since the epoch.

[MrGunflame]

This is very strange, I have no idea why cargo fmt would break formatting and then not fix it.

[Qqwy]

Hang on; I think I might be running a slightly older version of cargo fmt locally (from 2023-11) (My nightly version was up-to-date but my stable version wasn't). Let me see whether updating to the latest version fixes it.

[Qqwy]

That should've done it 👍

[MrGunflame]

Just a few leftovers from debugging it seems.

[Qqwy]

Just a few leftovers from debugging it seems.

Apologies, that was sloppy. I should stop trying to make these kinds of PRs late at night. Thanks for your keen eye!

[MrGunflame]

No need to apologize, thanks.

[MrGunflame]

Released in version 1.0.3.