should encode and decode both take an optional reference to a buffer

[d-v-b]

in numcodecs, Codec.decode takes an out parameter, which is an optional buffer-like for storing the result of decoding. encode does not, which seems strange, because this means encode will always allocate new memory (someone correct me if I'm off base here).

in zarr-python, neither our encoding or decoding routines take an out parameter, which I think entails that every codec allocates memory every time it encodes / decodes?

so I have 2 questions:

• should we change the zarr-python codec API to allow an out parameter for the decode functions?
• should we do the same thing for the encode functions?