Add nightly-optimizations feature

[davidhewitt]

Add a feature to gate the sequence implementations using buffer + specialization behind feature flag.

Questions:

• Should this feature be disabled or enabled by default?
• How should we configure CI to test this feature?

TODO:

• I'd like to write some documentation about this feature before merge. Wanted to propose this feature before I spend too long writing docs.

[kngwyu]

Thank you, but I don't think we should do this now since we can't test that we can compile the code without #![feature(specialization)].

[davidhewitt]

That's ok. I'm happy to leave this PR to wait until we have finished removing specialization from PyClass.

Also I have some other ideas I am exploring with ManagedPyRef / ToBorrowedObject to remove specialization there, which will be the only other stuff left

[davidhewitt]

FYI I continued to look at ManagedPyRef and ToBorrowedObject. I would propose that we remove ManagedPyRef for now (see #930) and add nightly-optimizations feature to ToBorrowedObject's specialization.

I will leave this here for now until the provider pattern to remove specialization from PyClass is implemented. After that, we should be able to test this PR fully - and compile without specialization!

[davidhewitt]

Just for completeness - my vision is that once we have made this feature gate, we can rework these optimizations to use min_specialization. And then we should eventually be able to get them onto stable Rust and remove the feature flag.

The feature flag will be a bridge which lets us compile on stable before min_specialization is stabilized.

[davidhewitt]

FYI I will rebase this after #961 is merged - and (fingers crossed) add a CI for stable rust at the same time.

[davidhewitt]

I will rebase this tonight unless you want me to close this and we focus on #969 ?

[kngwyu]

I'm sorry I close this PR in favor of #969