feat(tree-wide): migrate to Hyper 1.x

[liushuyu]

This pull request migrates the hyper crate to version 1.1.

I have to admit the situation is a bit disastrous due to the massive amount of API changes they have accumulated.

The following is a summary of the changes in this pull request:

• Due to hyper splitting the library. Some more crates were added to accommodate and retain the functionalities.
• Since reqwest (as of writing) still uses hyper 0.14, dependencies can not be shared anymore.
• hyper now requires the library user to provide the I/O interface, so you will find I have added an adapter/wrapper to avoid refactoring all the web-server-related functions to be async.
• The make_service macro is removed because the make_service_fn function was deleted. Instead, I added a taping-over function to stitch the services together like before.
• The shutdown handling is changed as hyper now delegates some event-loop responsibilities to the library user.

[Xuanwo]

The entire Rust ecosystem has undergone significant changes since that code was written. We're still using non-async crates like rouille and working directly with hyper, which increases our maintenance workload. Do you think it's a good opportunity for us to switch to axum or another framework?

[codecov-commenter]

Codecov Report

Attention: 49 lines in your changes are missing coverage. Please review.

Comparison is base (11703e0) 31.02% compared to head (c8f9c8b) 30.65%.

Files	Patch %	Lines
src/dist/client_auth.rs	0.00%	49 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2013      +/-   ##
==========================================
- Coverage   31.02%   30.65%   -0.38%     
==========================================
  Files          51       51              
  Lines       19617    19637      +20     
  Branches     9418     9428      +10     
==========================================
- Hits         6087     6019      -68     
- Misses       7874     7911      +37     
- Partials     5656     5707      +51     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[liushuyu]

The entire Rust ecosystem has undergone significant changes since that code was written. We're still using non-async crates like rouille and working directly with hyper, which increases our maintenance workload. Do you think it's a good opportunity for us to switch to axum or another framework?

axum is currently unstable (in terms of API). Other service frameworks within the direct tokio ecosystem are still trying to migrate to hyper v1.

So, I am unsure which approach is better given the current situation. The main issue as of now is still the architectural support. Ironically, non-mainstream devices usually need the caching the most (due to limited computing power), but sccache can't provide service for them.

[Xuanwo]

Thanks!