Decode via byte slice for memcache and file read

[bboreham]

This is more efficient, since the decoder can read field names in-place.
It also appears to be absolutely faster.

On master, decoding 75 reports from file, totalling 16MB compressed:

real	0m27.566s
user	0m26.432s
sys	0m1.312s

On this branch, calling NewDecoderBytes but not updating the msgpack libarary:

real	0m23.569s
user	0m22.764s
sys	0m1.172s

And, if you update the msgpack library so it actually does the in-place reading:

real	0m22.710s
user	0m21.820s
sys	0m1.328s

Note readReport decompresses rather than having ReadBytes do it, to avoid creating an intermediate copy of the entire compressed file.