fix(parser): support MySQL index options and prefix-length key parts in index DDL

[minleejae]

Refs #2490

Description

Fixes five of the six groups of MySQL index DDL reported in #2490. These now parse:

-- 1. key part with a prefix length *and* a direction (CREATE and ALTER)
CREATE INDEX i03 ON t (c1(20) DESC);
ALTER TABLE t ADD INDEX i05 (c1(20) DESC);

-- 2/3. CREATE INDEX index_option / algorithm_option / lock_option
CREATE INDEX i08 ON t (c1) KEY_BLOCK_SIZE = 8;
CREATE INDEX i10 ON t (c1) ALGORITHM = INPLACE LOCK = NONE;

-- 4. DROP INDEX algorithm_option / lock_option
DROP INDEX i15 ON t ALGORITHM = INPLACE LOCK = NONE;

-- 5. FULLTEXT / SPATIAL (previously fell back to UnsupportedStatement)
CREATE FULLTEXT INDEX i18 ON t (body) WITH PARSER ngram;
CREATE SPATIAL INDEX i19 ON t (g);

Two causes:

• IndexColumnWithParams() allowed only one CreateParameter() per key part, so (20) consumed it and the trailing ASC/DESC could not be matched. It now collects parameters in a loop. IndexColumnsWithParamsList() is shared by the CREATE and ALTER paths, so one change fixes both.
• The option keywords (KEY_BLOCK_SIZE, ALGORITHM, LOCK, NONE, PARSER, FULLTEXT, SPATIAL) were simply not reachable from the option lists that CREATE INDEX and DROP INDEX use.

No new grammar ambiguity

Those keywords all live in NonReservedWord(), so adding them as new alternatives to the trailing (...)* loops pushed JavaCC from 13 to 15 choice-conflict warnings, and explicit syntactic LOOKAHEAD did not help. They are instead added to the existing flat token lists of CreateParameter() and Drop(), which adds no new choice point and leaves the warning count unchanged.

LOCK is the one real ambiguity, since it also starts a LOCK TABLE statement. It is guarded so it is only taken as a DROP argument when it cannot begin the next statement:

LOOKAHEAD({ getToken(1).kind == K_LOCK && getToken(2).kind != K_TABLE })
tk=<K_LOCK> { dropArgs.add(tk.image); }

Scope

Group 6 (CAST(... AS UNSIGNED ARRAY) for multi-valued indexes) is not included: it needs an ARRAY marker on the public ColDataType model, and K_ARRAY_LITERAL is already used in six other places. Better as its own PR, so #2490 stays open.

Testing

• ./gradlew check passes (checkstyle, PMD, spotless, spotbugs, JaCoCo).
• 4923 tests, 0 failures, 0 errors.
• JavaCC reports 0 errors and 13 warnings, identical to master — verified by building both states and diffing the warning list.
• Seven new tests in CreateIndexTest, AlterTest and DropTest, including one that pins DROP TABLE t1; LOCK TABLE t2 IN SHARE MODE; still parsing as two statements.

Deparsing follows the existing conventions: c1 (20) DESC (matching the mycol2 (75) output already asserted in CreateIndexTest) and KEY_BLOCK_SIZE = 8. All statements above were verified against MySQL 8.4.11 while writing #2490.

PR Checklist

• I have read the contribution guidelines and the governance document on PR expectations.
• Minimal changes to code not directly related to your change
• One feature/change per PR unless tightly coupled
• Do a rebase on upstream master

[manticore-projects]

Much nicer now, thank you for work and effort!