Fix SQLite AUTOINCREMENT constraint ordering

[russellromney]

Summary

Fix SQLite generation for MySQL auto-incrementing primary keys so AUTOINCREMENT is preserved regardless of source constraint order and emitted in a SQLite-valid position.

Before this change, semantically equivalent MySQL definitions produced different SQLite SQL:

-- input
CREATE TABLE z (id INT AUTO_INCREMENT PRIMARY KEY)
-- previous SQLite output
CREATE TABLE z (id INTEGER PRIMARY KEY)

-- input
CREATE TABLE z (id INT PRIMARY KEY AUTO_INCREMENT)
-- previous SQLite output
CREATE TABLE z (id INTEGER PRIMARY KEY AUTOINCREMENT)

This also fixes a table-level primary key case that previously generated SQLite syntax rejected by SQLite itself:

-- previous SQLite output
CREATE TABLE x (id INTEGER NOT NULL AUTOINCREMENT PRIMARY KEY)
-- fixed SQLite output
CREATE TABLE x (id INTEGER NOT NULL PRIMARY KEY AUTOINCREMENT)

SQLite accepts AUTOINCREMENT after INTEGER PRIMARY KEY; placing it before PRIMARY KEY is a syntax error.

Changes

• Normalize SQLite column constraints so AutoIncrementColumnConstraint follows the column PrimaryKeyColumnConstraint when both are present.
• Preserve AUTOINCREMENT for AUTO_INCREMENT PRIMARY KEY as well as PRIMARY KEY AUTO_INCREMENT.
• Update MySQL and SQLite dialect tests to cover both source orders and the valid table-level primary key output.

Tests

python3 -m pytest tests/dialects/test_sqlite.py -q
python3 -m pytest tests/dialects/test_mysql.py -q

I also sanity-checked the fixed SQLite output with Python's sqlite3 module.

[VaggelisD]

Hey @russellromney, thank you for the PR! Just a couple of nits:

[VaggelisD]

This test seems like a duplicate of the other, I think I'd wrap it in a for loop e.g

for constraint in ("PRIMARY KEY AUTO_INCREMENT", "AUTO_INCREMENT PRIMARY KEY"):
  self.validate_all(
    ...,
    read={
      "mysql": f"CREATE TABLE z (a INT {constraint})",
   },
   ...
 )

[georgesittas]

Adding to this^, make sure to do with self.subTest if you do the loop.

[VaggelisD]

I think the previous for loop was a bit easier to understand plus we also don't have to loop over constraints twice with next; Should we revert and patch it with insert instead?

[russellromney]

All feedback addressed.