In the original issue [1] and commit [2], support for unexpected EOF was
added to the frontend to detect when a connection was closed abruptly.
Additionally, this allows us to differentiate normal io.EOF errors with
unexpected errors in the backend.
[1] https://github.com/jackc/pgx/issues/662/
[2] 595780be0f
When an message is received and a timeout occurs after reading the
header but before reading the entire body the connection state could
be corrupted due to the header being consumed. The next read would
consider the body of the previous message as the header for the next.
fixes#348
