test: pin the documented WL label form

[jdsika]

Follow-up to #63, from an adversarial review of it.

The gap

docs/api.md (lines 47 and 534) documents generated blank-node labels as
b followed by 12 hex digits, with an optional numeric collision suffix.
Consumers match on that shape rather than merely checking that labels changed -
linkml asserts this exact form in three separate modules to confirm its
--diff-stable flag actually reached the serializer:

• tests/linkml_runtime/test_utils/test_rdf_canonicalize.py
• tests/linkml/test_scripts/test_diff_stable.py
• .github/tests/test_diff_stable_install.py

Nothing in this repository defended that form. The wl tests check injectivity,
stability, cardinality and tie-suffix ordering, but never the label shape. So
widening the hash to 16 hex digits would have passed all 1259 tests, silently
falsified the documentation, and broken linkml - with the canary from #61 as the
first signal. That works, but it is an expensive way to discover that we changed
our own documented format.

Change

WL-LABEL-FORM in the requirement catalog, with two tests in
tests/wl/test_wl_contracts.py:

• test_labels_have_the_documented_form - every label is b + 12 hex digits,
  and an untied label is exactly 13 characters
• test_tied_labels_share_one_base_and_add_numeric_suffixes - tied nodes share
  one base and differ only by _1, _2, with the reported labels matching what
  wl_relabel_quads actually emits

Also

Two cleanups in #63's module, found in the same review: two tests took a
parse_format parameter they never used (ruff's selected rules E,F,W,B do not
flag unused arguments), and _reordered's docstring said "opposite order" while
the helper sorts descending.

Verification

1261 passed (2 new), coverage 96.01% against the 95% gate,
check_requirements.py verifies 50 supported requirements,
check_standards.py clean, ruff check src tests scripts clean.

No src/ change, so no behavior change, no emitted-byte change, and no
changelog entry or release.