Raise ProgrammingError immediately on concurrent connection use instead of blocking

[Copilot]

Using the same Connection object from multiple threads simultaneously is unsupported and leads to undefined behavior. Previously, the C-level lock used WAIT_LOCK, which silently serialized concurrent calls rather than surfacing the bug to callers. This change switches to NOWAIT_LOCK so that any second thread that attempts to use a connection already locked by another thread receives a ProgrammingError immediately.

C extension (_mysql.c)

• Unified lock implementation: free-thread and GIL builds both now use PyThread_type_lock (done in prior commit); the PyCriticalSection branch is gone.
• _mysql_ConnectionObject_Lock → int: uses NOWAIT_LOCK; sets ProgrammingError and returns -1 on contention.
• _mysql_ConnectionObject_LockWait (new): blocking WAIT_LOCK variant used exclusively in _mysql_ResultObject_dealloc, where mysql_free_result() must be called regardless.
• BEGIN_CONNECTION_OPERATION: propagates lock failure via return NULL before checking open.
• Four direct BEGIN_CONNECTION_LOCK call sites updated to check the return value and propagate errors correctly (handles both PyObject * and int return contexts).

Tests (test_connection.py)

Replaced the two "operations are serialized" tests with tests that assert concurrent access raises ProgrammingError immediately:

# Thread A: conn.query("SELECT SLEEP(0.5)")  ← holds lock ~0.5 s
# Thread B (100 ms later):
with pytest.raises(ProgrammingError, match="already in use"):
    conn.thread_id()   # must fail instantly, not block

Both Connection method contention and Result.fetch_row contention are covered.