Convert event offset to int, avoiding overflow with i32

[barakugav]

What does this implement/fix?

The function _compute_robust_event_table_position read the offset of the event tables. Since numpy 2.x.y, the following line returns np.int32 rather than pythonic int:

(event_table_pos,) = np.frombuffer(fid.read(4), dtype="<i4")

This causes an overflow in other places, such as

data_size // (n_samples * n_channels)

Additional information

[larsoner]

This is also in progress at #13548, can you see if that PR fixes your problem?

[barakugav]

This is also in progress at #13548, can you see if that PR fixes your problem?

Seems like it does, yes