Deps: prefer sub-crates of `futures`

[BastiDood]

This PR attempts to shorten compilation times by preferring dependencies on the sub-crates of futures rather than the entire futures crate itself. The changes I've made practically slim down the default features (of futures) to just two crates: futures-util and futures-channel.

To make CI pass, I also needed to fix some Clippy warnings. Most of them were mechanical changes.

With that said, I don't expect any breaking changes. At the very least, there should be a slight improvement in compilation times. Thanks! 🎉

[sfackler]

How much does this improve build times?

[BastiDood]

System Information

The following measurements have been tested on a laptop with the following specs: Rust Version	OS	CPU	RAM
1.63.0 (Stable)	Windows 10 21H2 (19044.1889)	Intel Core i7-10750H @ 2.60GHz (12 CPUs)	8192 MB

I then ran the following commands and recorded the (cold) compilation times reported by Cargo:

cargo clean                # clean up old artifacts
cargo build -p postgres    # only compile `postgres` on debug

cargo clean                # clean up old artifacts
cargo build -p postgres -r # only compile `postgres` on release

Native Windows

On native Windows, the build times improved ever so slightly on debug builds. For the release builds, however, I must say that the improvements may still be within the realm of noise.

However, there is an improvement in the number of crates needed to compile. Before the PR, a Windows build required 105 crates. Now, it only requires 103 crates. This is likely due to the absence of the futures-executor crate in the production dependencies.

Version	Debug	Release
Before PR	27.38s	34.71s
After PR	25.06s	34.58s

Linux

I also measured on an Ubuntu 22.04.1 build (via Windows Subsystem for Linux). There is also a slight improvement on debug builds. Release builds remained more or less the same.

Regarding the number of crates compiled, the Linux side had two fewer crates to compile (from 101 to 99). This difference is consistent with native Windows. Again, this is likely due to the absence of the futures-executor crate.

Version	Debug	Release
Before PR	17.20s	23.52s
After PR	16.46s	23.29s

Conclusion

There is a slight improvement in cold debug builds. In both Windows and Linux (via WSL), we reduced the number of compiled crates by two. For projects with larger dependency trees, removing futures-executor from our direct dependencies will allow Cargo to compile postgres earlier in the topological sort—perhaps even in parallel with futures-executor!