Avoid all-null concat warning in coverage regression fixture #461

The local full gate for #460 failed with 5,080 passing tests and one FutureWarning in a new coverage test fixture. The product checks passed; the fixture constructing duplicate unnamed-version requests used pd.concat between an all-None object column and a string column. With pandas inferred strings enabled, that construction emits the all-NA concatenation deprecation.

Fix the fixture by constructing the intended records as one DataFrame, and run it explicitly under both object-string and inferred-string modes. Keep warnings as errors; no warning filter or production workaround is needed. Link the fix from #460 and rerun the full one-worker gate.