feat(client, server): remove the tcp cargo feature and related code

[MrGunflame]

Removes the tcp cargo feature and related code from server and client. I've kept client::connect even thought it doesn't serve much use without tcp-related features.

Closes#2856.

[seanmonstar]

Woah, awesome work, thanks!

[MrGunflame]

It should be noted that this change currently breaks pretty much every test and example, which all rely on the Server/Client type.

[seanmonstar]

Alright, the meta pieces are in place now, so full-speed ahead on the feature work...

Hm, it looks like we'll need some sort of local utility for the unit tests...

[MrGunflame]

Yes, I think all tcp feature code is gone now. I've been working on converting the tests to use conn client/server mod, but I'm not quite happy with it. The examples also pose a similar problem.

[seanmonstar]

I'm looking through the tests right now, and it seems like a good time to clean up this mess anyways 😅 . Most of the client test! stuff can probably have the inner part of the macro just use conn and TcpStream instead of Client. The conn impl tests are probably fine as-is.

... As I scroll more, I'm not sure there's many tests that are checking the higher-level Client specifically... If they are, we can comment them out and move them to the https://github.com/hyperium/hyper-util crate as part of the Client move.

[seanmonstar]

Aww, sorry, it looks like merging #2896 caused conflicts with this 🙈

[MrGunflame]

That's fine, I'll have to update the examples anyways before this works.

[Robert-Cunningham]

When we merge this PR, will Exec struct in common/exec.rs have a purpose anymore? Or will we then want to replace Exec everywhere Box<dyn Executor> (and simultaneously add TokioExecutor to hyper-utils)?

[MrGunflame]

I think you have an argument there to replace Exec with an Arc<dyn Executor>, but I'll likely just keep change the Exec enum into a struct (and remove the Default variant).

[MrGunflame]

All tests except get_strip_connection_header should pass now. For some reason get_strip_connection_header gets stuck when using the proxy connection and I haven't been able to find out why that is yet.

[seanmonstar]

Thanks for plugging away at this! I'll kick off another CI run.

[MrGunflame]

I believe it should finally pass now.

[seanmonstar]

Looks like a test or benchmark needs to be updated. I know this has grown big, if you want to just comment out the tests that are failing with a TODO to re-enable them, we can merge and file follow-up PRs to get the remaining bits. Whatever you'd find easier and more satisfying <3

[MrGunflame]

Yeah, I totally forgot that nightly and benches exist. I can probably rewrite them, but I'd need some more time for that. If the PR is blocking other changes I can of course comment out the benches and add them in later again.

[seanmonstar]

There's some other items that are somewhat separate, but I think they will cause merge conflicts with this, so getting it merged sooner than later will probably mean less pain. For instance, I was about to start deleting hyper::Server and hyper::server::accept, but realized that'd cause conflicts with this. But up to you, there's other things to work on too.

[MrGunflame]

Alright, I will just comment them out for now I guess, might be able to add them in later in another PR or work on some other changes then.

[seanmonstar]

Thanks for taking this so far! I've squashed and fixed up the last couple conflicts and merged in #2929.