[stateless_validation] test_chunk_forwarding_optimization fails with --features statelessnet_protocol

[jancionear]

Runing the integration test test_chunk_forwarding_optimization with --features statelessnet_protocol causes the test to fail:

$ cargo nextest run --package integration-tests --features statelessnet_protocol test_chunk_forwarding_optimization
...
test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 367 filtered out; finished in 2.66s


--- STDERR:              integration-tests tests::client::features::access_key_nonce_for_implicit_accounts::test_chunk_forwarding_optimization ---
thread 'tests::client::features::access_key_nonce_for_implicit_accounts::test_chunk_forwarding_optimization' panicked at integration-tests/src/tests/client/features/access_key_nonce_for_implicit_accounts.rs:779:9:
assertion failed: PARTIAL_ENCODED_CHUNK_FORWARD_CACHED_WITHOUT_HEADER.get() > 0.0

It looks like the PARTIAL_ENCODED_CHUNK_FORWARD_CACHED_WITHOUT_HEADER isn't increased as it should.

The test passes with --features nightly and no special features (stable).

The test was modified in #10527 to make it work with stateless validation. It started passing with --features nightly and the fix was merged. Now it turns out that it doesn't work with --features stateless_protocol, so it'll need to be adjusted further.

It should be fixed quickly because it causes CI runs to fail: (e.g https://github.com/near/nearcore/actions/runs/7872120842/job/21476689267)

Refs: #10506, zulip conversation

[jancionear]

Ah, the check for increased metric is disabled on nightly beacause "we're so efficient with part forwarding":

   // With very high probability we should've encountered some cases where forwarded parts
   // could not be applied because the chunk header is not available. Assert this did indeed
   // happen.
   // Note: For nightly, which includes SingleShardTracking, this check is disabled because
   // we're so efficient with part forwarding now that we don't seem to be forwarding more
   // than it is necessary.
   if !cfg!(feature = "nightly") {
       assert!(PARTIAL_ENCODED_CHUNK_FORWARD_CACHED_WITHOUT_HEADER.get() > 0.0);
   }

That sounds weird. But I guess we can also disable the check for feature = "statelessnet_protocol" as a quick fix 0_o
/cc @Longarithm

[jancionear]

The check was disabled for "nightly" in #10582 by @robin-near, 4 days ago.
Can I assume that the same reasoning can be applied to statelessnet_protocol, which also uses SingleShardTracking?
It sounds reasonable to me, I'll open a PR.