Flaky near-head calibnet tests

# Summary

Some tests like `EthGetBlockTransactionCountByNumber` check `Latest` or `Pending`, which is inherently flaky - it assumes two nodes are on the exactly same head which might not be the case.

This fails some runs, e.g., https://github.com/ChainSafe/forest/actions/runs/32744137240/attempts/1?pr=7531

We need to do it somehow smarter; either make somehow sure both nodes are exactly on the same head at the time of the call (IMO pretty challenging to achieve in practice), tinker with the predicate (medium), or just relax the success condition to `basic` (easy). If we go with the `basic` path, make sure that we not only test `latest`/`pending` but some less tipsets as well.

# Completion Criteria

- [ ] reduce flakiness of RPC parity tests on `latest`/`pending` tags.

# Additional Links & Resources