Add basic benchmark reporting to CI

This enables keep alive on tcp sockets via keep-alive feature.

The choices were based on the request_policy tip from the COMMAND
output.

Trying to avoid bumping MSRV to 1.63 just for a
dev dependency.

Fixes redis-rs#539, redis-rs#703

This caused panics when the node was in the wrong state.

This will give us more data on test failure.

rustls/rustls#1383

…dis-rs#923)

Fixes redis-rs#922.

This adds ~10 minutes to previous test run time, in order to leave a
comfortable margin, while still significantly reducing the run time of
the task in case it gets stuck. Without defining this, the task runs
for 6 hours.

* Release redis 0.23.2 / redis-test 0.2.2

* Update rust version for CI, due to tokio update.

* Update rust-version in cargo.toml.

* Fix lints.

The extern crate statement used to be an important part of rust, several years ago.

These days it's a point of confusion for newcomers without adding any value, and should be omitted.

The test was broken by version 1.0.105 of serde_json, due to
serde-rs/json#1055 .

* Add NOTBUSY error code.

* async cluster: Group responses by response_policy.

This change implements the response_policy tip in the async cluster.
https://redis.io/docs/reference/command-tips/#response_policy

This means that the results from fan-out commands will be aggregated
differently, based on the sent command.

* Determine the type of a key

* modify the name of the function for determining the type of a key

* add unit test for cmd type

* unit test for key_type function

* edit unit test test key_type after cargo fmt --all -- --check

---------

Co-authored-by: MARONE <[email protected]>

This fixes a bug where 2 servers use the same port, because
`get_random_available_port` is called before the first server manages to
use its port. This should reduce test flakyness.