fix(postgres): treat FULLTEXT and SPATIAL as unreserved identifiers

[fmguerreiro]

Problem

PostgreSQL schemas that use fulltext as a column name (common with tsvector/trigram search) failed to parse:

CREATE TABLE public.film (
    fulltext tsvector NOT NULL
);

Error: SQL parse error: sql parser error: Expected: a data type name, found: 'fulltext'

Root Cause

parse_optional_table_constraint consumed the FULLTEXT token before realising it was not a valid constraint keyword for the current dialect, then backtracked with prev_token(). The consume-then-backtrack pattern left the parser in a subtly broken state for non-MySQL/Generic dialects, causing the next parse_column_def_inner call to see the wrong current token.

Fix

Add a peek-based early-return guard in parse_optional_table_constraint before any token is consumed. For dialects other than MySQL/Generic, FULLTEXT and SPATIAL are valid identifiers and must not be touched by the constraint parser:

if name.is_none()
    && self
        .peek_one_of_keywords(&[Keyword::FULLTEXT, Keyword::SPATIAL])
        .is_some()
    && !dialect_of!(self is GenericDialect | MySqlDialect)
{
    return Ok(None);
}

Test

Added parse_fulltext_column_and_index_in_postgres in tests/sqlparser_postgres.rs covering:

• CREATE TABLE film (fulltext TSVECTOR NOT NULL) — column named fulltext with tsvector type
• CREATE INDEX film_fulltext_idx ON film USING gist (fulltext) — GiST index on a fulltext column

Version

Bumped to 0.60.13.