Update validation for graphml parser + tests

[Brqndon1]

Summary

Added input validation to parse_graph_attributes in GraphmlParser, mirroring the validation style of fact_parser.py but emitting warnings.warn(...) and skipping the offending attribute (or whole node/edge) instead of raising.

Validation rules implemented

1. Node/edge ID validation against _COMPONENT_RE. One warning per bad node/edge, all attributes on it skipped (not one warning per attribute).
2. Attribute key validation against _PREDICATE_RE.
3. Bound parsing — fixed the int(...) → float(...) bug in the comma-separated interval string parser. Added range and order checks (0 <= lower <= 1, 0 <= upper <= 1, lower <= upper).
4. Single-numeric out-of-range values (and out-of-range numeric strings like "1.5") now warn-and-skip instead of silently falling through to a categorical label.
5. Composed key-value labels are validated against _PREDICATE_RE — values containing spaces, commas, parens etc. now warn-and-skip rather than producing malformed labels.
6. Empty / whitespace-only keys and values are rejected.

Tests

New file tests/unit/dont_disable_jit/test_graphml_parser.py mirroring the structure of test_fact_parser.py:

• TestValidGraphmlParsing (10 tests) — numeric in range, single-string numeric, interval string (regression), edge with numeric, edge with interval, categorical labels, boundary values 0 / 1 / 0.0 / 1.0, multiple valid attributes, static_facts=True propagation.
• TestInvalidGraphmlParsing (12 tests) — every bullet from the task spec's "Invalid" list: bad keys, empty keys, out-of-range numerics (string and type), bad node/edge IDs, all four interval-string failure modes, composed-label regex failures, whitespace values.
• TestEdgeCasesConditions (4 tests) — empty graph, node with no attributes, label accumulation across multiple nodes, mixed-case keys preserved.