Fix COPY FROM and add tests

[drdrsh]

COPY commands handling had a bug that manifests with high pool contention or very show idle_timeouts.

When copying data from client to a server, you first send a COPY query. Something like

COPY copy_test_table FROM STDIN CSV

You get back a CopyInResponse (G) message from the server. Client is then expected to send zero or more CopyData (d) messages to the server and end with CopyComplete (c) or CopyFailed (f) messages.

In transaction mode, PgCat allows the server connection to be released back into the pool after it gets G message from the server. In high contention situations or with very short idle timeout, the client's attempt to checkout another connection results in the client sending CopyData packets to a server connection that is not in copy mode.

This PR fixes this bug by keeping track of copy_mode and avoiding the release of the connection back into the pool if it is in copy_mode.

[levkk]

Awesome!