Add link about Python 2 GZip workaround [ci skip]

Lest we or others wonder what this workaround is doing. Link back to the
PR comment for details.

TL;DR - GZip on Python 2 (and Python pre-3.5) determines the number of
bytes included via `len`, which is needed for the GZip footer. This will
be incorrect if the data is not flattened and using a single byte type.
Hence we cast to `uint8` here to satisfy this constraint. On Python 3.5+
this is not needed as it casts the data to a `memoryview` and accesses
the `nbytes` attribute to determine this.

ref: #128 (comment)

Author: jakirkham

numcodecs/gzip.py

@@ -29,6 +29,7 @@ def encode(self, buf):
         buf = ensure_contiguous_ndarray(buf)
         if PY2:  # pragma: py3 no cover
             # view as u1 needed on PY2
+            # ref: https://github.com/zarr-developers/numcodecs/pull/128#discussion_r236786466
             buf = buf.view('u1')
 
         # do compression