I spent some time checking which parts of c-kzg are not tested by the test vectors. I did this by exporting coverage information of the C code of c-kzg when the test vectors are tested through the Rust bindings. I was forced to go through the Rust bindings because only the bindings run the test vectors at the moment (and IMO that's fine).

For what it's worth, I used the cargo-llvm-cov project and the following command:
CC=clang LLVM_COV=llvm-cov LLVM_PROFDATA=llvm-profdata cargo llvm-cov --include-ffi --html.
In case you are curious, I attach a zip file (c_kzg_coverage.zip) with the coverage report of c-kzg (commit f5e4b030f) when running the tests of the Rust bindings.

All in all, the coverage results were very good (i.e. the test vectors are well thought out!).

In this PR, I fix some minor issues found during the above procedure:

• I add a missing cell_id range check in recover_all_cells() (which was actually tested by the test vectors)
• Add a test vector for too many cells in recover_all_cells() which represents the right side of the spec assertion: assert CELLS_PER_EXT_BLOB / 2 <= len(cell_ids) <= CELLS_PER_EXT_BLOB. The test vector is sorta rendundant because the combination of the duplicate cell_id check and the cell_id range check would have caught this edge case, but still worth checking explicitly since it's not trivial (and it shows up in the coverage as well).

Some more notes:

• This c-kzg check shows up as unvisited in the coverage report, but it's not covered by the spec. We should either add it to the spec, or remove it from c-kzg?
• This c-kzg check shows up as unvisited in the coverage report, but it's actually tested by the function's caller. Perhaps it should be an assert? Similar to the sanity check in recover cells?