build: bump minimum zarr to >=3.1.6

[TomNicholas]

Follow-up to #1023 (merged), fixing the CI failure its xfail was papering over.

The real bug

#1023 added test_parser_recurses_into_subgroups and marked it xfail on zarr < 3.1.1, on the theory that zarr 3.1.0 couldn't list nested groups. That was wrong. Investigating the min-versions CI failure showed the actual issue:

zarr < 3.1.6 mis-strips keys when listing the contents of a nested group — e.g. an array stored at group_a/nested/deep_var is reported as a member named var, which isn't a real object, so it's silently dropped and array_keys() returns empty. It triggers when array names share substrings with others in the store (root_var/a_var/deep_var), the same class of bug as zarr-python#3657 that test_parser_with_nested_store_path already works around.

Verified across versions (array stored at group_a/nested/deep_var):

zarr	array_keys() of group_a/nested
3.1.0 – 3.1.5	❌ [] (array dropped)
3.1.6+	✅ ['deep_var']

It is not version-of-zarr-specific in the depth sense, and it is not fixable by choosing tidy test names — real user data can have any names, so on any zarr 3.1.0–3.1.5 a hierarchical store with colliding names silently loses nested arrays. The honest fix is to require the zarr version that fixed it.

This PR

• Bump zarr>=3.1.0 → zarr>=3.1.6 (and the minimum-versions pin to ==3.1.6). 3.1.6 is the exact version that fixed the listing bug.
• Remove the unnecessary xfail (and HAS_NESTED_GROUP_LISTING) from test_parser_recurses_into_subgroups, keeping deliberately colliding array names so the test is a real regression guard. It passes at 3.1.6 and fails at 3.1.5 (confirmed locally).
• Re-include the tiff and grib parsers in the minimum-versions env — they were excluded only because they require zarr>=3.1.2 / >=3.1.1, both satisfied by the new floor.

CI watch

The minimum-versions job now resolves virtual-tiff and gribberish against zarr 3.1.6 — confirm that environment still solves cleanly. If either fails to resolve at the minimum pins, the fallback is to drop it back out with an accurate reason.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 90.01%. Comparing base (2cc8ff3) to head (d299d31).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1024      +/-   ##
==========================================
- Coverage   90.10%   90.01%   -0.10%     
==========================================
  Files          36       36              
  Lines        2204     2204              
==========================================
- Hits         1986     1984       -2     
- Misses        218      220       +2     

see 1 file with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.