Make array field adapter independent from the connection.

[stas]

Since the field does not depend on the database OIDs, calling prepare() is not necessary. This is also mentioned in the docs1. The reason this is not dependent on the connection since we don't try to adapt any special OIDs that can vary from one Postgres version to another.

We discovered this when a wrapped connection was used to adapt the new type.
Relevant: opbeat/opbeat_python#68

Thanks in advance for reviewing this.

[stas]

Hmm, it looks like there's something strange with the way py3 handles the unicode string adapter.

This could be related to psycopg/psycopg2#331
I've tried a couple of things, including calling adapted.prepare(conn) and explicitly setting the connection encoding to UTF8, still on py3 I get the adapter result as bytea.

We can close this until the py3 support is not ready... :|