Session config lost if error is returned on first query after two successive reconnects

[msullivan]

Here, I set a session config variable. Then, I twice run a query that errors (here, select asdf), while restarting edgedb between commands so that the CLI needs to reconnect. If I do this twice in a row, the session config gets lost! If I only do it once, then it manages to keep the config.

This is probably somewhat marginal for end users, but I hit it in my development workflow because I'm constantly running queries that don't work and then restarting the server to run a hopefully fixed version!

_localdev:edgedb> configure session set apply_access_policies := false;
OK: CONFIGURE SESSION

_localdev:edgedb> select cfg::Config.apply_access_policies;
[false]

_localdev:edgedb> select asdf;
Reconnecting...
error: InvalidReferenceError: object type or alias 'default::asdf' does not exist

_localdev:edgedb> select asdf;
Reconnecting...
error: InvalidReferenceError: object type or alias 'default::asdf' does not exist

_localdev:edgedb> select cfg::Config.apply_access_policies;
[true]

[msullivan]

I'm not sure if this case was a regression or if this was always fiddly (see https://github.com/edgedb/edgedb-cli/pull/1183)

[msullivan]

Aha! And it also loses the automatically set session_idle_transaction_timeout! I've been haunted by session idle transaction timeouts for years, and I had still been getting it even after we were pretty sure we had squashed it in the server!

_localdev:edgedb> select cfg::Config.session_idle_transaction_timeout;
{<duration>'0:00:10'}
_localdev:edgedb> start transaction;
OK: START TRANSACTION
_localdev:edgedb[tx]> 
Reconnecting...
_localdev:edgedb>