CHECK detector treats quoted column names as constraint expressions

The raw schema scanner can invent CHECK violations or reject valid CHECK expressions because it does not skip backtick/bracket identifiers correctly.

Reproduced at f5d4636e70:

```sql
CREATE TABLE t(id INTEGER PRIMARY KEY, `CHECK(0)` INT);
INSERT INTO t VALUES(1,1);
SELECT dolt_verify_constraints('--all');
SELECT * FROM dolt_constraint_violations;
PRAGMA integrity_check;
```

Actual: verify returns 1 and records a violation on t, while integrity_check returns ok. The same schema and row produce zero violations in Dolt. A divergent merge using this schema also reports invented constraint violations.

A real `CHECK(`a)b`>0)` on a column named `a)b` additionally causes verification to fail parsing the expression.

Replace the character scanner with parsed CHECK expressions or SQLite tokenization. Add equivalent-schema tests across supported identifier quoting styles, with keywords and parentheses inside identifiers.

Source: https://github.com/dolthub/doltlite/blob/f5d4636e70/src/doltlite_merge_constraints_check.c#L9

Start in src/doltlite_merge_constraints_check.c at the raw schema scanner, then run the provided CREATE TABLE, INSERT, verification, violations query, and integrity_check reproduction. Compare behavior for backtick and bracket identifiers, including keywords and parentheses inside names. Done means equivalent schemas produce no invented violations and real CHECK expressions such as CHECK(`a)b`>0) verify without a parse failure, with tests covering the supported quoting styles.

Written by the indexing model from the issue text.