fix: recognize BEGIN READ WRITE/ONLY as transaction statements

The statement splitter checks if a keyword after BEGIN is a transaction
keyword to distinguish transaction statements from PL/SQL blocks.  READ,
ONLY and WRITE were missing from this list, causing Redshift's
BEGIN READ WRITE to be treated as a block start — split() would return
the entire content as one statement.

Add READ, ONLY, WRITE to the transaction keyword list so
BEGIN READ WRITE and BEGIN READ ONLY are correctly split as standalone
transaction statements.

Fixes: #843
Signed-off-by: Vincent Gao <gaobing1230@gmail.com>

Author: binggao1230

sqlparse/engine/statement_splitter.py

@@ -132,7 +132,8 @@ def _change_splitlevel(self, ttype, value):
                 (ttype is T.Keyword or ttype is T.Name) and \
                 unified in ('TRANSACTION', 'WORK', 'TRAN',
                             'DISTRIBUTED', 'DEFERRED',
-                            'IMMEDIATE', 'EXCLUSIVE'):
+                            'IMMEDIATE', 'EXCLUSIVE',
+                            'READ', 'ONLY', 'WRITE'):
             self._seen_begin = False
             if self._block_stack and self._block_stack[-1] == 'BEGIN':
                 self._block_stack.pop()