pgconn: apply a 10-second connection timeout everywhere

[brmzkw]

PGCONNECT_TIMEOUT=10 is injected into every subprocess environment so that pg_dump, pg_dumpall, pg_restore, and psql all time out quickly when a packet-dropping firewall is in the way. connect_timeout=10 is added to the DSN so the same limit applies to Go-side Open/Ping calls via pgx.

The timeout is hardcoded (not configurable) and intentionally only covers the initial TCP handshake and authentication, not the duration of a running dump or restore.

[brmzkw]

On OSX, tested with the following:

Run a container:

docker run --rm -ti --name test -p 9999:5432 -e POSTGRES_PASSWORD=postgres postgres:17

Drop connection:

 echo "block drop out proto tcp to any port 9999" | sudo pfctl -ef -

Run a backup:

plakar -stdio at ./toto backup postgres://postgres:postgres@localhost:9999
/Users/niluje/dev/plakar/plakar/plakar: failed to backup source: rpc error: code = Unknown desc = querying server version: failed to connect to `user=postgres database=postgres`:??[::1]:9999 (localhost): tls error: write tcp [::1]:50815->[::1]:9999: write: socket is not connected??127.0.0.1:9999 (localhost): dial error: timeout: dial tcp 127.0.0.1:9999: connect: operation timed out??[::1]:9999 (localhost): failed to write startup message: write failed: write tcp [::1]:50818->[::1]:9999: write: socket is not connected??127.0.0.1:9999 (localhost): dial error: timeout: dial tcp 127.0.0.1:9999: connect: operation timed out

Revert firewall:

sudo pfctl -F rules
sudo pfctl -f /etc/pf.conf