Support dialect-specific CREATE INDEX definitions

[minleejae]

CREATE INDEX rejects Spanner's UNIQUE NULL_FILTERED combination, SQL Server's UNIQUE NONCLUSTERED combination, and the schema-qualified PostgreSQL collation/operator-class example in #1994. This change parses and preserves these definitions with an explicit dialect:

• SPANNER: keep uniqueness in Index.getType() and null filtering in CreateIndex.isNullFiltered().
• SQLSERVER: reuse Index.getClustering() independently of uniqueness.
• POSTGRESQL: accept qualified collation/operator-class names and bare function keys. Store key attributes once in their structured fields so AST edits are reflected in SQL output.

Share key/option expression rendering between the AST and deparser, and route CREATE INDEX through the common definition traversal used by visitors, table discovery and validation. Preserve the default parser's single-type and raw key-parameter behavior, including MySQL prefix lengths.

Fixes #652. Addresses the index example in #1994 and the CREATE UNIQUE NONCLUSTERED INDEX failure in #1563. This does not resolve all failures in the Sakila scripts linked from #1563. A regression test also covers a PostgreSQL function followed by an index and another statement.

Validation:

• Gradle check: 6,272 tests, no failures, 25 skipped; formatting and static checks passed. The local formatter ratchet uses the upstream base commit because the fork's origin/master is behind upstream.
• Maven verify passed.
• Coverage includes dialect isolation, modifier combinations, AST edits, expression visitors/deparsers, validation and statement boundaries.

Syntax references: Spanner, SQL Server, PostgreSQL.

[manticore-projects]

thank you!