Lift the alias and quoting traits into the vocabulary layer

[trueqbit]

Stacked on #1552.

Summary

The open-work item "lift the remaining traits into the vocabulary layer" of docs/internals/vocabulary-layer.md, for everything below ast/: the alias traits and those of table references, table-valued expressions and literals were declared and defined in their own headers, alias_holder and alias_column_t had no trait, and consumers pattern-matched them - along with column_pointer.

• alias.h, column_pointer.h, literal.h and table_reference.h move to dev/ast/.
• Declared in grammar_traits_fwd.h (new "Alias traits" section): is_alias, is_column_alias, is_recordset_alias, is_table_alias, is_cte_moniker, is_table_valued_expression, and the new is_alias_holder and is_alias_column. Declared in structural_traits_fwd.h (quoting): is_literal, is_table_reference. The alias traits stay defined in alias_traits.h next to alias_tag, since aliases are user-definable rather than sqlite_orm nodes; the others are defined right after their nodes.
• Consumers now matching by trait: node_tuple (column alias and alias holder in one, table-valued expression), column_result_t (alias column, column pointer), statement_serializer (alias holder, column alias, alias column), ast_iterator, cte_storage.h, cte_column_names_collector.h, table_name_collector.h, storage_impl.h (materialize_column_pointer(), find_column_name()), the multi ORDER BY streaming in serializing_util.h, and unpacked_arg in ast/app_function.h.
• The column pointer referring to a CTE column alias used to be a partial specialization refining the one for column_pointer<T, F> (in column_result_t, and as overloads in storage_impl.h) - which is why they matched the template rather than a trait. They are now told apart by the column pointer's field_type being an alias holder.
• Fix: the type trait is_column_alias derived from is_alias, and thus held for record set aliases as well, unlike is_column_alias_v. It now agrees with the variable template - which affects as<A>(), explicit CTE columns and the ast_iterator skipping aliases, none of which is meant to see a record set alias.
• The open-work item is closed. Neither remaining is_*_v belongs in the traits layer: is_rtree_table_element_or_constraint is an admissibility composite and moves beside its base and FTS5 siblings in vocabulary/algorithms/ddl_predicates.h, and is_dbo_name_literal_v recognizes a reflection annotation that never becomes a node, so it stays in schema/dbo_name.h. The guidance on which is_*_v qualify moves to "Deciding where something goes".

Test plan

• tests/static_tests/expression_classification.cpp: new "aliases and quoting" section - the alias traits including is_column_alias against a record set alias, is_alias_holder, is_alias_column, is_literal, is_table_reference, is_table_valued_expression, and the node tuples of a column alias and an alias holder.
• The CTE, alias and column pointer static tests and the CTE storage tests exercise the rewritten column pointer paths.

🤖 Generated with Claude Code