Retry docker compose up to ride out Docker Hub pull flakes

[klagrida]

Problem

Transient Docker Hub pulls (registry-1.docker.io ... context deadline exceeded) were failing compose jobs outright (e.g. analyst-ci this week) and needed a manual re-run. With #61 they now fail fast — but still fail.

Change

Wrap all 16 docker compose … up invocations across the 11 compose workflows in a bounded 3-attempt retry (15s backoff).

for i in 1 2 3; do
  if docker compose … up …; then break; fi
  if [ "$i" = 3 ]; then echo "still failing after $i attempts"; exit 1; fi
  echo "retrying …"; sleep 15
done

Correctness

• The compose command sits in an if condition, so it's exempt from the runner's set -e — a failed attempt falls through to the retry instead of aborting the step.
• The 3rd failure exit 1s, so a genuinely-broken run still fails (a naive for … && break loop would return the trailing sleep's 0 and silently mask failure — avoided here).
• The first attempt runs the identical command, so a healthy job's behaviour is unchanged.

Verified: all 11 workflows parse; bash -n clean; retry semantics tested under bash -eo pipefail (success→0, total-failure→1, fail-then-succeed→0). Only up commands are wrapped — logs/down untouched.

🤖 Generated with Claude Code