Tokio version 1.23 breaks the tests

[boraarslan]

Upgrading the Tokio version 1.23 breaks the tests: CI Run

Tests pass without a problem on macOS and some Linux distros but fail on other instances.

My first instinct was this PR being responsible: tokio-rs/tokio#5223 since on the test:
test_mailbox_send_with_backpressure_counter_no_backpressure_cleansheet we measure some backpressure on the zero capacity channels where we expect none:

        FAIL [   0.002s]                 quickwit-actors mailbox::tests::test_mailbox_send_with_backpressure_counter_no_backpressure_cleansheet

--- STDOUT:                              quickwit-actors mailbox::tests::test_mailbox_send_with_backpressure_counter_no_backpressure_cleansheet ---

running 1 test
test mailbox::tests::test_mailbox_send_with_backpressure_counter_no_backpressure_cleansheet ... FAILED

failures:

failures:
    mailbox::tests::test_mailbox_send_with_backpressure_counter_no_backpressure_cleansheet

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 48 filtered out; finished in 0.00s


--- STDERR:                              quickwit-actors mailbox::tests::test_mailbox_send_with_backpressure_counter_no_backpressure_cleansheet ---
thread 'mailbox::tests::test_mailbox_send_with_backpressure_counter_no_backpressure_cleansheet' panicked at 'assertion failed: `(left == right)`
  left: `15`,
 right: `0`', quickwit-actors/src/mailbox.rs:466:9
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

However, inserting tokio::time::sleep(Duration::ZERO).await; before sending the second message fixes the test. I am not sure what causes this, but my guess is context switch causes the task to resume after some time; in the meantime, the BackPressureActor processes the message, and the channel becomes empty again.