v2/v3 differences for the crc32c codec

[d-v-b]

In V2, the crc32c codec takes a location parameter (either "start" or "end") which determines whether the checksum is prepended or appended, respectively. In a regression, the zarr v3 definition of the crc32c codec doesn't take a location parameter and instead only appends.

In the interest of not keeping around two marginally different crc32c codecs that do the exact same thing, we should think about how to design one codec class that can handle the fact that, in the zarr v3 context, the codec must be configured to append, while in a zarr v2 context, the codec can append or prepend.