SoC: Target postget wrappers

[thorwhalen]

The control of what pandas reader to use (role played by df_from_data_given_ext through ext_specs argument should be generalized and made more generally reusable. Proposed steps:

Target the creation of postget functions for wrap_kvs where data is assumed to be binary and key->deserializer function specification is separated and made more easy to specify. Right now, only (data, ext) controls the deserialization. That's a particular case of (data, key) which we don't want to rule out.
For example, the name of a file could indicate two different ways of treating .csv (one with a header, one without).

Offer a few out of the box stores (with LocalBinaryStore, FilesToZip, ...)