fix(gen-shacl): enforce operator exactness, nested-slot parity, numeric bounds

[jdsika]

Summary

Resolves every substantive finding from the adversarial audit of the
rule-converter stack (see the audit comments on #19, #20, #21). One commit,
21 regression tests; 20 fail on the pre-fix source (the 21st locks in
already-correct zero-member has_member semantics that had no coverage).

Audit finding	Fix
A1/B2/C2 — greedy operator dispatch (conditions mixing recognized + unrecognized operators, expression-level any_of/all_of/none_of/exactly_one_of, co-set equals_string + equals_string_in, value_presence: ABSENT combined with a bound — all partially translated, dropping conjuncts → demonstrated false positives)	New _set_operator_fields enumerates the constraint-bearing fields actually set on a condition / class expression (metadata excluded; scalars never judged by truthiness, so minimum_value: 0 still counts). Named patterns and the fallback dispatch only when their conditions set exactly the operators they translate; everything else is skipped.
A2 — boolean-guard shadows PIV (equals_string: "true" on a string-range slot hijacked into a boolean comparison, flagging conforming data)	Boolean guard dispatches only when the target slot's induced range is boolean; otherwise the rule flows to presence-implies-value, which compares the string.
B1/C1 — inner slots resolved against the outer class (member-class slot_usage slot_uri overrides diverged from sh:path → constraints silently never fired; name collisions with outer-class overrides made FILTER NOT EXISTS vacuously true → false positives)	_member_conditions resolves the container slot in the outer class's induced context, requires its range to be a class, and resolves inner slot URIs and enum values against that range class. Container slot_usage range-narrowing now picks the right enum too (_resolve_member_enum_ref subsumed by passing the range class to _resolve_enum_value_ref).
B3/C3 — non-numeric threshold bounds ("abc" → unparsable sh:select poisoning the whole shapes graph; YAML date → arithmetic 2020-01-01 = 2018, silently never firing; .nan, booleans)	_sparql_number returns None for anything but int / finite float (bool excluded) and callers skip the rule. Docstring corrected (the metamodel range is Anything, not int/float).
B4 — elseconditions silently dropped	Rules with an else branch log a warning that only the forward direction is enforced, consistent with the bidirectional/open_world warnings.
C4 — alias-form / unknown rule keys (a key my_slot for a slot named my slot, or a typo'd key, fabricated a default_prefix predicate no shape uses → silently vacuous constraint, or an always-firing has_member)	New _rule_slot resolves rule keys via the class's induced slots, then the underscored alias form, then the base slot; _slot_uri returns None for unresolvable names and every converter skips the rule.
Test gaps (M5 zero-member untested; vacuous "<Manual>" not in query assertion)	Zero-member has_member end-to-end test added; assertion now checks the full IRI form.

Areas of uncertainty

All substantive audit findings, including C4 (alias-form / unknown rule keys),
are addressed here. \b/\f escaping was verified a non-issue (both are
legal raw characters in SPARQL STRING_LITERAL2).

Stack position

main
└─ feat/shaclgen-rules-sparql                      #11 (green)
   └─ feat/shaclgen-presence-implies-value-stacked   #19
      └─ feat/shaclgen-compositional-rule-fallback   #20
         └─ fix/shaclgen-rule-converter-hardening    #21
            └─ fix/shaclgen-rule-converter-audit-findings ← this PR

When upstreaming, fold each fix into the submission that introduces the
affected code (exactness + boolean gate → framework/PIV; nested parity +
numeric gate → fallback). These fixes also apply verbatim to the consolidated
branch of #18 (cherry-pick b9287ac1).

How was this tested?

• tests/linkml/test_generators/test_shaclgen.py: 127 passed (106 base
  • 21 new; per finding: structural skip assertions, generated-query IRI
    assertions, and pyshacl end-to-end with advanced=True).
• Negative control: with this branch's tests against the pre-fix
  generator source, 20 of the 21 new tests fail — the tests bite.
• ruff check and ruff format --check clean on both changed files.

Checklist

• My code follows the contributor guidelines
• I have added tests that prove my fix/feature works
• Existing tests pass locally with my changes

AI Assistance

If you used AI tools while preparing this PR, you are still the author and responsible for understanding, verifying, and defending your submission. Please engage with reviewers personally rather than through your agent during feedback and revisions. See our AI Covenant for details.