Allow "encoding" and "errors" attributes to be updated at runtime

[andymccurdy]

This patch provides three changes:

1. Adds the Reader.set_encoding method. This method accepts the "encoding"
   and "errors" arguments, verifies that both are either NULL or found within
   the codecs module and then updates the Reader's state to use the new values.

   The motivation here is to allow a user to update the Reader's encoding
   functionality at runtime, perhaps temporarily. redis-py plans to use this to provide
   context manager support. Something like:

   client = redis.Redis(encoding="utf-8")
   # gets the value of the "my-utf8-key" decoded as "utf-8" and returns a unicode string
   client.get("my-utf8-key")
   
   with client.response_context(encoding=None) as no_decode_client:
       # gets the value of "my-bytes-key" and returns it as a bytes object
       no_decode_client.get("my-bytes-key")

   In the above example, the context manager will call Reader.set_encoding,
   passing the encoding=None argument to update the Reader state. When the
   context manager exits, it will again call Reader.set_encoding to restore the original "encoding" and "errors" values.

2. Allows both Reader.__init__ and Reader.set_encoding to accept Python None
   values for "encoding" and "errors". When encoding=None is specified,
   hiredis-py will not decode strings and instead return bytes objects.

3. Passing an invalid "encoding" to Reader.__init__ or Reader.set_encoding
   will immediately raise a LookupError. Prior to this, Reader.__init__ would
   accept an invalid encoding and then raise the LookupError the first time
   the Reader attempted to decode a string. Raising the error immediately
   seems preferable.

[ifduyue]

Do we need to expose reading encoding, using tp_getset, tp_descr_get or some other methods?

>>> print(reader.encoding)
'utf-8'
>>> reader.encoding = 'ascii'
>>> reader.encoding
'ascii'

Overall looks good to me.

[andymccurdy]

I’m fine exposing read access to those attributes if you’d like, but it’s not something I need.

I generally don’t like setting attributes that can raise exceptions, but that’s just a stylistic preference. I’ll do whatever you recommend. E.g, reader.encoding = 'unknown' would raise an error.

[ifduyue]

Let’s keep it this way for now, I’ll have another look at it

[ifduyue]

Here is a reference count issue, better use Py_RETURN_NONE

[andymccurdy]

@ifduyue Good catch! Made that change.

[ifduyue]

Thanks, merged!