Improve query runner error handling

[seanlinsley]

This PR updates the query runner to submit the first error it sees, to avoid reporting an unhelpful current transaction is aborted error back to the server. BEGIN READ ONLY was a late addition to #628, so this wasn't caught in manual testing.

This also fixes an error I saw while testing this, where server.QueryRuns is nil:

panic: assignment to entry in nil map

goroutine 99 [running]:
runner.connect.func3()
    runner/websocket.go:150 +0x4ac
created by runner.connect in goroutine 64
    runner/websocket.go:113 +0xa44

[keiko713]

In #641 (comment), you mentioned that this is executed even if the firstErr is not nil for the case of first one failed with a timeout. However, even though the cause was a timeout, the second one will fail regardless:

pgaweb=# set statement_timeout = 1000;
SET
pgaweb=# begin read only;
BEGIN
pgaweb=*# select pg_sleep(2);
ERROR:  canceling statement due to statement timeout
pgaweb=!# select pg_sleep(0.5);
ERROR:  current transaction is aborted, commands ignored until end of transaction block

Meaning this code doesn't quite do what you want it to do. When the firstErr is not nil (for whatever the reason), this second query execution will always be ignored.

[seanlinsley]

Good catch. Maybe the query text should contain BEGIN READ ONLY and COMMIT so each query is in its own transaction.

[keiko713]

Nice, confirmed that it's reporting the correct error message 👍