fix(gfql/binder): replace regex WHERE label narrowing with AST-based analysis (#1125) - #1193

…analysis (#1125)

The binder's _apply_where_label_narrowing() previously fell back to regex
(_WHERE_LABEL_RE / _WHERE_NON_CONJUNCTIVE_RE) to extract label narrowing
from raw WHERE expression text. This was fragile: a quoted string such as
WHERE n.name = 'n:Admin' could produce false-positive label assignments, and
the OR/NOT/XOR guard was text-level and sensitive to unusual expressions.

Root cause: the Lark grammar resolves the ambiguity between
`where_predicates` and `expr` in favour of `expr` for AND-joined label
predicates like `WHERE n:Admin AND n:Active`, so they reached
generic_where_clause as raw text rather than as structured AST.

Fix (two-part):
- parser.py: extend generic_where_clause to recognise AND-joined bare label
  predicates via _BARE_LABEL_PREDICATE_RE and produce structured
  WhereClause.predicates, eliminating the need for any regex in the binder.
  Non-label parts (property comparisons, OR/NOT/XOR) fall through to the
  raw-expr path and are conservatively not narrowed.
- binder.py: remove _WHERE_LABEL_RE, _WHERE_NON_CONJUNCTIVE_RE, and the
  regex fallback block; _apply_where_label_narrowing now derives label
  narrowing exclusively from WhereClause.predicates (AST).

New tests: triple AND conjunction, XOR (no narrowing), quoted label pattern
(no false positive), string-with-colon (no false positive), nested OR-AND
expression (no narrowing). All 1416 GFQL tests pass.

Co-Authored-By: Claude Sonnet 4.6 <[email protected]>

…ntry

Co-Authored-By: Claude Sonnet 4.6 <[email protected]>