Re-introduce covariance for containers

[sosthene-nitrokey]

MpMc and spsc::Queue are not migrated to the new pattern because they were already invariant due to the UnsafeCell.

Fix #501

[sosthene-nitrokey]

Rebased. CI is failing only because of CHANGELOG.

[newAM]

Thanks!

This looks logical, but I'm not confident enough to merge this on my review alone, will wait for another review from @rust-embedded/libs to merge.

[jannic]

It doesn't mean much because I'm not a member of the libs team, and also don't know much about the internals of heapless.

But I was curious and spent some time to understand both the issue and the proposed solution. The key information not obvious from other parts of the documentation was https://rustc-dev-guide.rust-lang.org/variance.html#variance-and-associated-types. After I found that, I'm convinced your proposal is a good solution. It's unfortunate that it complicates the code a bit, but it's probably worth it.

I tried to measure the effect on binary sizes and compiled the embassy uart_r503 example with heapless 0.8.0 and with this branch, and the version from this branch was actually 20 bytes smaller than with 0.8.0.
Notably, I was unable to compare it with the current main branch, because that didn't compile: It failed with some entirely unhelpful error message about lifetimes. :-) I guess this shows that the fix is indeed useful.

[therealprof]

Looking great to me.