Share one SQL tokenizer between SqlTableReference and LibsqlResultSetMetaData #111

#70 grew SqlTableReference.java from 266 to 1149 lines. The file is now over 1000 lines.

LibsqlResultSetMetaData keeps a second near-identical tokenizer for the auto-increment check from #56. It has its own token record, tokenize, skipQuoted, unescape, and isWordChar. Any tokenizer defect exists twice.

Inside the parser, matchingParen and matchingParenIn, identifiers and identifiersIn, and indexOfWord and indexOfWordIn duplicate each other.

Expected:

• Extract one SqlTokenizer with a token type, a depth, and a literal kind.
• Use it from the parser and from the CREATE TABLE scanner.
• Remove the duplicated helpers.
• Do this after the parser defect issues, so the fixes land on one implementation.