identifier_case: keep the case of backtick-, acute- and bracket-quoted identifiers

[dylanpulver]

Addresses the case-mangling half of #433.

IdentifierCaseFilter exempts only "..." from case conversion, but SQL_REGEX recognizes four quoting styles for identifiers. The other three — `x` (keywords.py:132), ´x´ (:133) and [x] (:173) — arrive as plain T.Name and get converted:

>>> sqlparse.format('select * from `MyTbl`', identifier_case='upper')
'select * from `MYTBL`'

That breaks the statement on MySQL with lower_case_table_names=0 (the Unix default, docs) and on SQL Server under a case-sensitive collation — and backticking the name is precisely the workaround suggested on #433. test_identifiercase_quotes already pins the intended behaviour for "..."; this extends it to the other three. sqlparse/utils.py:58 already counts ` as a quote character, so the filter was the odd one out.

Ran: make test → 515 passed on 3.10/3.11/3.12/3.13/3.14 (unmodified master in the same session: 506 passed; the 9 extra are the new parametrized cases). make lint clean — it was clean before the change too.

Two mutants, same command and env: reverting only filters/tokens.py fails 9/9 of the new cases; a naive fix that exempts just the backtick still fails 6/9. So the test does discriminate between remedies rather than merely passing.

Alternative I did not take: make the lexer emit T.String.Symbol for all four styles, which is where the asymmetry originates. That changes a public token type, so it is breaking and belongs on its own.

Not covered: dialects where a delimited identifier is case-insensitive anyway — there this is a no-op rather than a fix.

How I found it: a systematic audit of sqlparse's tables, diffing the quoting styles SQL_REGEX recognizes against the ones each filter knows about. Not from hitting it in production — weigh it accordingly. #433 turned up afterwards and matches.

---

• ran the tests (pytest, via make test)
• all style issues addressed (ruff check sqlparse/)
• your changes are covered by tests
• your changes are documented, if needed (CHANGELOG)

AI disclosure: prepared with assistance from Claude Opus 5 (claude-opus-5). Every command and number above I ran and read myself.