Optimize event extension with return_array=True flag

[janreising]

In Events.get_extended_events(return_array=True) astroCAST creates an array of size num_events * num_frames. This is only efficient if the events are extended over the full num_frames, but creates an array with mostly np.nan values if a fixed extension window is chosen.

if memmap_path:
                # todo only size that we need?
                memmap_path = Path(memmap_path).with_suffix(f".dtype_{np.dtype(dtype).name}_shape_{n_events}x{n_frames}.mmap")
                arr_ext = np.memmap(memmap_path.as_posix(), dtype=dtype, mode='w+', shape=(n_events, n_frames))
            else:
                arr_ext = np.zeros((n_events, n_frames), dtype=dtype)

A more efficient solution would be to get the maximum event length after extension and create an array of this size instead.