Fix orphaned CI droplets on cancelled runs

[benvinegar]

Problem

CI droplets leak when a GitHub Actions run is cancelled mid-create. The droplet gets created on DigitalOcean but the ID never makes it to GITHUB_OUTPUT, so the cleanup step has nothing to destroy.

Fix

• Tag all CI droplets with baudbot-ci on creation
• Pass the deterministic droplet name (ci-{distro}-{run_id}) to the cleanup step as a third arg
• cmd_destroy falls back to looking up the droplet by name when no ID is available
• Add droplet.sh list command for debugging orphans

Still needed

The 10 existing orphan droplets need to be manually destroyed. Run:

export DO_API_TOKEN=...
# These predate the baudbot-ci tag, so list all:
curl -s -H "Authorization: Bearer $DO_API_TOKEN" \
  "https://api.digitalocean.com/v2/droplets?per_page=100" \
  | python3 -c "import json,sys; [print(f'{d[\"id\"]}  {d[\"name\"]}') for d in json.load(sys.stdin)['droplets']]"
# Then destroy each:
for id in <ids>; do
  curl -s -X DELETE -H "Authorization: Bearer $DO_API_TOKEN" \
    "https://api.digitalocean.com/v2/droplets/$id"
done