RFC: Drop COOKIE auth support

[zeenix]

COOKIE auth is not needed at all on Unix platforms, has a complicated implementation (especially the server side) and drags in sha1 dep (see #530). The only real use case I see for it is Windows, where we currently don't support EXTERNAL auth but once that's fixed (#723), we could think about this.

There is also the use case of TCP but I'm not sure, that's worth keeping either. If UDS is available on all platforms (that we care about), there is no good reason to use TCP. For remote use case, TCP should be highly discouraged since it's unencrypted and COOKIE auth won't work for that anyway.

Together with we can #726, we can really simplify the auth code and pipeline the client-side handshake better (making it go much faster).

[zeenix]

Dropping the idea in favor of #731.

[zeenix]

I'm rethinking this. @elmarco do you think it'd be a big deal?