Call pgconn_set_internal_encoding_index in all branches of pgconn_set_default_encoding

[drdrsh]

This PR reintroduces a call to pgconn_set_internal_encoding_index in pgconn_set_default_encoding. This call was removed in version 1.3.0 (specifically in this PR).

When Encoding.default_internal is set, ruby-pg sends a SET client_encoding TO ... query, if that query succeeds everything works great. If it fails and the connection is closed, well nothing breaks.

Now assuming Ruby and Postgres are both using UTF-8 encoding, if that query fails AND the connection remains open, we end up with desynchronized encoding between Ruby and Ruby-pg where Ruby's encoding is UTF-8, Postgres Encoding is UTF-8 but Ruby-pg falls back to SQL_ASCII (I think?) causing it to encode the incoming UTF-8 strings from Postgres into ASCII.

Version 1.2.3 handles encoding query failure gracefully. The following output shows this behavior in action in versions 1.2.3, 1.5.3, and a patched version of 1.5.3

[1.2.3 Response] [{"user_id"=>"1", "username"=>"™™™™™™"}, {"user_id"=>"2", "username"=>"™™™™™™"}]
[1.5.3 Response] [{"user_id"=>"1", "username"=>"\xE2\x84\xA2\xE2\x84\xA2\xE2\x84\xA2\xE2\x84\xA2\xE2\x84\xA2\xE2\x84\xA2"}, {"user_id"=>"2", "username"=>"\xE2\x84\xA2\xE2\x84\xA2\xE2\x84\xA2\xE2\x84\xA2\xE2\x84\xA2\xE2\x84\xA2"}]
[1.5.3-fix Response] [{"user_id"=>"1", "username"=>"™™™™™™"}, {"user_id"=>"2", "username"=>"™™™™™™"}]

This repo has a docker stack that reproduces that behavior and demonestrates the fix.

[larskanis]

Thank you very much for providing patch and repro example! This is part of pg-1.5.4 which I just released.

[drdrsh]

Thank you for the quick turnaround, Lars!