- We maybe shouldn't be in charge of closing
- Actually have a functioning implementation
- Split ReadFragmentHeader off from ReadFragment

Given a bunch of Readers, read the first fragment header from each;
check that they all have the same backend, fragment size, and original
data size; and return an ErasureCodeParams instance with an appropriate
backend name and an upper-bound for the number of parity fragments.

Currently, the caller needs to know the appropriate number of data
fragments -- in the future we might be able to guess this from the ratio
of original data size to fragment size.

You need to provide the number of data fragments, but take a bunch of
fragment archives and try to re-create the original file.