fix: INNER of a following INNER JOIN wrongly consumed as table source alias (#6711)

[CuriousLinYu]

What does this PR do?

Fixes #6711.

When a FROM clause contains a join followed by another INNER JOIN:

SELECT c.id
FROM context_table c
INNER JOIN loan_table l ON c.id = l.id
INNER JOIN adjust_table a ON l.id = a.id

the INNER token of the second join is silently consumed as the alias of the previous table source / join node, producing a corrupted AST:

• the first join node gets alias = "INNER"
• the following join degrades to a plain JOIN
• serialization emits invalid SQL with a bogus AS INNER:

SELECT c.id
FROM context_table c
INNER JOIN loan_table l ON c.id = l.id AS INNER
JOIN adjust_table a ON l.id = a.id

Since the corrupted output is still accepted by the same parser, re-validation with the same parser cannot detect the semantic damage.

Root cause

SQLSelectParser.parseTableSourceRest() guards LEFT / RIGHT / FULL against being consumed as a table-source alias when they actually start the next join (lookahead for JOIN / OUTER / ANTI / ARRAY / SEMI), but INNER was missing from that lookahead group, so it fell through to the alias path.

Fix

Add case INNER with the same lookahead as LEFT / RIGHT / FULL, so INNER is only treated as an alias when it is not followed by JOIN / OUTER / ANTI / ARRAY / SEMI. Fully symmetric with the existing handling; real INNER aliases (when not followed by a join keyword) keep working.

SQLSelectParser is the shared base parser, so this affects all dialects that reuse it (mysql, odps, postgresql, ...) — not only odps as reported in the issue.

Tests

New test class Issue6711 covering:

• INNER JOIN followed by INNER JOIN (odps / mysql / postgresql) — asserts both join nodes are INNER_JOIN with null aliases
• LEFT / RIGHT / FULL JOIN followed by INNER JOIN
• three consecutive INNER JOINs
• parse → print → parse → print round-trip stability, and output must not contain AS INNER
• single INNER JOIN regression guard

Verified locally: mvn -pl core test passes with the full core module suite (1759 test classes, 0 failures, 0 errors).

Before the fix, 5 of the 6 new tests fail on master (commit fa8dc99); after the fix all 6 pass.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

lintyu seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

lintyu seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}