client encoding name normalization in psycopg 2.4

Psycopg 2.4 now tries to map the PostgreSQL client encoding to a Python
codec.  But it fails to consider some variant spellings.  For example, I
have

PGCLIENTENCODING=UTF-8

in my environment, which completely breaks the new psycopg version with

no Python codec for client encoding 'UTF-8'

(The underlying reason for this is that I use

export PGCLIENTENCODING=$(locale charmap)

which gives variant spelling of encoding names across operating
systems.)

The PostgreSQL backend normalizes an encoding name by removing all
non-alnum characters from it.  I suggest psycopg do the same.

Attached is a patch that implements that.  Note that the PostgreSQL
backend version of this actually lowercases the encoding names during
normalization.  I have made this patch uppercase them to keep the patch
smaller, but you may want to consider doing the lowercasing, to keep
things consistent.