Increase test coverage for statement-distribution

[Sophia-Gold]

We should run code coverage on statement-distribution and add more tests for branches that aren't already covered.

[mrcnski]

Coverage results

I managed to get some results with cargo llvm-cov:

https://app.codecov.io/gh/mrcnski/polkadot-sdk/blob/master/polkadot%2Fnode%2Fnetwork%2Fstatement-distribution%2Fsrc%2Fv2%2Fmod.rs (ignore the gum lines, don't know how to disable those.)

The results are worrying... There are some big untested paths, even entire untested functions:

• handle_grid_statement
• handle_incoming_acknowledgement. I confirmed this manually, looks like we test sending BackedCandidateKnown but never receiving it.
• Interestingly, we also don't cover the line causing issues in statement-distribution: v2: Reduce reputation CostMinor("Unexpected Statement") #2346, identified by @alexggh.

This is using LLVM instrumentation which is supposed to provide the most accurate profiling.

Adding coverage for repo?

The profiling chokes on wasm, I guess that's why we don't use it. AFAICT it must be run on a per-crate basis. For example, it works on pvf-common but not on pvf, which has rococo-runtime as a dependency.

It does seem useful to run manually in targeted situations.

Instructions

These are the commands I ran:

cargo install cargo-llvm-cov
# runs `cargo test` with coverage profiling and opens report
cargo llvm-cov --open

To upload to codecov:

1. Make a fork of polkadot-sdk
2. Find the fork on codecov.io and follow instructions for setting up "Other CI"
3. Run:

cargo llvm-cov --codecov --output-path codecov.json
bash <(curl -s https://codecov.io/bash) -f codecov.json

[mrcnski]

I also ran profiling for prospective-parachains and the coverage there is really good! We could just add a test for the remove_candidate function.