Update CONTRIBUTING.md with proper LOOM test invocation

[adelavegaf]

Loom tests currently fail when LOOM_MAX_PREEMPTION is set to 1. Specifically, yield_calls_park_before_scheduling_again across all schedulers (current thread, multi thread, multi thread alt).

For example:

RUSTFLAGS="-Dwarnings --cfg loom  -C debug_assertions" LOOM_MAX_PREEMPTIONS=1 LOOM_MAX_BRANCHES=10000 cargo test --lib --release --features full -- --nocapture loom_current_thread
    Finished release [optimized] target(s) in 0.05s
     Running unittests src/lib.rs (/Users/ant/Developer/oss/tokio/target/release/deps/tokio-32b7cb1c593730d4)

running 3 tests
test runtime::tests::loom_current_thread::assert_no_unnecessary_polls ... ok
thread 'runtime::tests::loom_current_thread::yield_now::yield_calls_park_before_scheduling_again' panicked at tokio/src/runtime/tests/loom_current_thread/yield_now.rs:22:17:
assertion `left == right` failed
  left: 1
 right: 0
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
thread 'runtime::tests::loom_current_thread::yield_now::yield_calls_park_before_scheduling_again' panicked at /Users/ant/.cargo/registry/src/index.crates.io-6f17d22bba15001f/loom-0.7.1/src/rt/execution.rs:216:13:
deadlock; threads = [(Id(0), Blocked(Location(None))), (Id(1), Blocked(Location(None))), (Id(2), Blocked(Location(None)))]
thread 'runtime::tests::loom_current_thread::yield_now::yield_calls_park_before_scheduling_again' panicked at /Users/ant/.cargo/registry/src/index.crates.io-6f17d22bba15001f/loom-0.7.1/src/rt/thread.rs:276:39:
called `Option::unwrap()` on a `None` value
stack backtrace:
   0:        0x104dfbfe8 - <std::sys_common::backtrace::_print::DisplayBacktrace as core::fmt::Display>::fmt::h3be43c6214e69043
   1:        0x104e19a70 - core::fmt::write::h12f9049830901a82
   2:        0x104df9bbc - std::io::Write::write_fmt::h2ff514dfe799c882
   3:        0x104dfbe1c - std::sys_common::backtrace::print::h722d6ff0889d39c2
   4:        0x104dfd958 - std::panicking::default_hook::{{closure}}::hd7f9985c1b9ae0aa
   5:        0x104dfd6a0 - std::panicking::default_hook::h29fb35819c2a32ca
   6:        0x104dfdecc - std::panicking::rust_panic_with_hook::h401df0236a0e7dae
   7:        0x104dfdc5c - std::panicking::begin_panic_handler::{{closure}}::h67f4ee7322151c3c
   8:        0x104dfc46c - std::sys_common::backtrace::__rust_end_short_backtrace::hbce128a2ff88406e
   9:        0x104dfda10 - _rust_begin_unwind
  10:        0x104e3192c - core::panicking::panic_fmt::hecc1dfe94cb15bc7
  11:        0x104e319b4 - core::panicking::panic::h73b079eecd9000fe
  12:        0x104e318b0 - core::option::unwrap_failed::h2a1d2a89129dbc81
  13:        0x104cd4880 - loom::rt::object::Ref<T>::set_action::h606daf31b2ba97a6
  14:        0x104cadf58 - scoped_tls::ScopedKey<T>::with::h634eac6aaf7668ce
  15:        0x104cb6d64 - loom::rt::atomic::Atomic<T>::store::he9584501adecba00
error: test failed, to rerun pass `--lib`

Caused by:
  process didn't exit successfully: `/Users/ant/Developer/oss/tokio/target/release/deps/tokio-32b7cb1c593730d4 --nocapture loom_current_thread` (signal: 10, SIGBUS: access to undefined memory)

Although the test consistently fails with this invocation (tried on my Mac and Linux using the repo's main branch), it's not failing on CI.

When looking into how we invoke loom in CI, I noticed we pass LOOM_MAX_PREEMPTIONS=2 (e.g. https://github.com/tokio-rs/tokio/actions/runs/8814945156/job/24195889256). This env var change makes the test pass.

Motivation

Loom tests fail when following the contribution guide on the main branch, despite them succeeding in CI. This is confusing for new comers like me!

Solution

This change updates the sample invocation provided in the contributing guide to match what we do in CI to ensure all loom tests pass.

[Darksonn]

This is a bug in the test, and ideally we fix the test.

[Darksonn]

Closing in favor of #6537.

We don't want to change the command, because increasing LOOM_MAX_PREEMPTIONS makes the tests take a lot longer.