The Zarr v3 codecs don't apply any input normalization. That means the following thing happens:
>>> from numcodecs.zarr3 import LZ4
>>> LZ4().codec_config
{'id': 'lz4'}
>>> LZ4()._codec.get_config()
{'id': 'lz4', 'acceleration': 1}
I can see no reason to make these different.