Use Prom2 chunk code

Much of the code was converted in #555, but currently Cortex uses chunk code from Prometheus v1.
In v2, chunks are broadly encoded in the same way, with a less-convoluted data structure.  However the Go APIs are rather different.

We'd need to still be able to read v1 chunks for a year minimum, so some wrapping/indirecting layer is required.
