Refuse to run against postgres with a non-C collation

[richvdh]

A number of people have reported (#6696) database index corruption which basically stems from their databases having a UTF-8 collation setting.

Our documentation tells you to set the collation to "C" to avoid this sort of thing (and possibly improve performance, according to postgres-general), but this is little comfort to people who missed that advice and now have corrupt indexes to deal with.

By way of a preventative measure, maybe we should complain if somebody is using a different collation?

synapse=# SELECT datcollate FROM pg_database WHERE datname = current_database();
 datcollate 
------------
 C
(1 row)

[richvdh]

#6734 addresses this