test(core): freeze the columnar-versus-object wire measurement

The wire freeze's closed list asked for this number since the beginning and
it had never been run: columnar form against an array of objects, on frozen
repositories, with the design fixed before the results.

The object arm reuses the codec's own emitter parts rather than restating
them, so it cannot quietly become a different encoder; its bytes transpose
back to encode_canonical_json byte for byte. Both positive controls hold --
one predicts the padded delta to the byte and caught a defect in itself, the
other puts the probe's resolution at 64 KiB so the reported gaps are
readings rather than noise.

The verdict it reaches is INCONCLUSIVE and stays so. The ingest oracle
refused eighteen of the twenty-one frozen repositories, so the three that
survived are two parts this repository's own material and one external
project. That is an instrument proof, not a population.

Author: orenlab