feat(index): support flattened JSON sub-doc indexing

[wirybeaver]

Parent: #4516
Draft PR: #7377

Motivation

Lance JSON inverted indexes should preserve array/object structure so path queries over arrays of objects can distinguish exact positions, wildcards, and nested-object constraints.

Tantivy documents a known flaw of bag-of-terms JSON array indexing: a query like cart.product_type:sneakers AND cart.attributes.color:red can match even when sneakers and red live in different objects in the same array. Lance's current single-token-stream JSON index has the same class of problem.

Proposal

Introduce FlattenedSubDocs mode for new JSON inverted indexes. A JSON row can flatten into multiple internal sub-documents, and matches are mapped back to the original Lance row id. This lets related JSON path constraints match inside the same flattened array element before row-id remapping.

Also add disable_cross_array_unnest=false by default, with an opt-in memory-safety tradeoff for documents with many sibling arrays, following Pinot's DisableCrossArrayUnnest behavior.

Expected behavior

• Existing JSON indexes remain readable using SingleDocument mode.
• New JSON indexes use FlattenedSubDocs mode.
• Positional and wildcard queries can return different row sets, for example foo[0].bar[0],str,y vs foo[0].bar[*],str,y.
• Nested-object constraints should be expressed as one JSON MatchQuery with semicolon-separated triplets plus Operator::And, so matching happens at flattened sub-doc granularity.
• Separate BooleanQuery.must child queries remain row-level composition and are not the query shape for nested-object semantics.