Fix dx serve proxying of websocket connections established from the browser

[arvidfm]

This PR makes the following changes:

• Check for the Upgrade header to detect websocket connections where the URI doesn't include the scheme
• Replace redirect with explicit proxying of websocket connections

The explicit proxying is required as browsers (at least Firefox and Chrome) don't handle redirect responses. E.g. Chrome emits an error complaining about an invalid response.

This PR is somewhat of a proof of concept in its current state. I'm happy to make any necessary changes to conform to Dioxus's code style and general guidelines, and to discuss alternative approaches.

Fixes #3894

[arvidfm]

@ealmloff Hi, just wondering if there's anything I need to/can do here to help move this along? If you're not interested in the change I'm happy to close the PR, just let me know whichever way.

[ealmloff]

@ealmloff Hi, just wondering if there's anything I need to/can do here to help move this along? If you're not interested in the change I'm happy to close the PR, just let me know whichever way.

Thanks for the ping. I think there were some conflicts with #3820. If you can fix those issues, this looks good to me once checks pass. I confirmed this works with axum websockets and with the server function websocket integration.

Requesting Jon's review on this as well since he wrote the original websocket proxy logic

[arvidfm]

I fixed the compilation errors following the Axum upgrade. Unfortunately even though Tungstenite and Axum now use the same Utf8Bytes type for its messages there doesn't seem to be an easy way to reuse the allocation since Axum doesn't expose the underlying Tungstenite Utf8Bytes value that it wraps. You can in principle do something like Bytes::from(axum_string).try_into().unwrap() but I was hesitant to introduce an unwrap() call even if it's guaranteed to succeed in this case.

Not that I expect it'll actually have a measurable impact on performance, it just feels a bit silly.

[arvidfm]

It's not entirely clear to me what the cause of the Playwright issues is, but I don't think it's related to these changes? From what I can tell digging through the logs it looks like the dioxus crate is failing to build for certain tests due to the specific combination of feature flags passed

[ealmloff]

The playwright test failures are not related. Playwright is failing on main. #4042 fixes some of the issues