ci: fix smoke and e2e workflows for the multi-instance model

[radnov]

The smoke and e2e workflows were still running the old single-instance flow, so they broke on every push after the multi-instance rework in #99. This gets them green again.

I moved both onto the new flow - generate the stack envs, create an instance, then bring up Traefik, monitoring and the instance under a ci project - and scoped every docker compose call to that project. The pytest suite made the same single-instance assumptions, so I pointed the helpers and tests at the per-instance project, loaded the instance env for the admin credentials, and looked up the Prometheus and Tempo containers by their compose labels instead of a bare docker compose exec.

A couple of things surfaced while getting the e2e run to pass. clean-all only tore down the app project, so I made it drop the Postgres volumes and per-instance network too, which gives the backup/restore test a genuinely fresh database. restore was passing its file variables through sudo, which strips them, so now they go via docker compose run -e the same way backup already does it. And I taught the login helper to retry the whole login when Chromium hits ERR_NETWORK_CHANGED from Docker churning bridges mid-test.

I also added development-2.0 to the workflow triggers so CI runs on this branch and PRs into it, rather than only once it merges up to master.