fix(ci): make every PostgreSQL readiness loop fail closed

[NikolayS]

Severity

Medium — CI reliability and diagnosability during release hardening.

Problem

Six PostgreSQL startup steps still use this pattern:

for i in $(seq 1 30); do
  docker exec ... pg_isready ... && break
  sleep 1
done || { echo "PG not ready"; exit 1; }

The done || ... handler is ineffective: if PostgreSQL never becomes ready, the final sleep 1 succeeds, so the loop itself exits successfully. The setup step goes green and a later install/test step fails with a less actionable connection error and without container logs.

Affected jobs in .github/workflows/ci.yml:

• PostgreSQL matrix
• frozen stable smoke
• Python client
• Go client
• TypeScript client
• Ruby client

The upgrade, pg_tle, pg_cron, and pg_timetable jobs already use an explicit ready flag and fail closed, so behavior is inconsistent within the same workflow.

Expected fix

• Convert every remaining loop to an explicit success flag (or a tested shared helper).
• Probe the actual target database with psql -c 'select 1' where initialization completion matters, rather than accepting the temporary init server.
• On timeout, print docker logs <container> and exit nonzero in the startup step.
• Add a lightweight shell regression/static assertion so future jobs cannot reintroduce the broken done || form.

This is the incomplete remainder of the readiness work discussed around #286.