Decide whether a claimed `fixed_size_binary(16|32)` should reshape rather than diverge between planners

The two DataFrame planners answer the same frame differently for a
fixed_size_binary(16) or (32) column carrying a claim it cannot fit: the
Arrow capsule path writes BINARY, and the NumPy planner refuses the column
outright. 24 cells of test/claim_matrix.py record this, and
./proj.py grid claim prints them on every run.

The divergence is inherited, not new: binary(16) with no claim at all
already splits the same way, which CHANGELOG.rst:48-55 documents as a
deliberate decision.

Making the claimed case reshape to object would close it, but would then
leave claimed and unclaimed fixed-size-binary behaving differently on the same
planner. That trade belongs to whoever made the original call rather than to a
review loop.

Asked for: an owner, and the outcome recorded either way — "we keep the
divergence, here is why" is a fine answer, and the claim grid then pins it as
intended behaviour rather than as an open question nobody re-reads.

Start with test/claim_matrix.py and run ./proj.py grid claim to inspect the 24 cells for fixed_size_binary(16) and (32). Read CHANGELOG.rst:48-55 for the documented binary divergence and compare the Arrow capsule and NumPy planner outcomes. Done means an owner records the decision and the claim grid clearly pins the behavior as intentional.