Records, per run, which worker cluster (cell) and lane (`container` or
`microvm`) the most recent attempt executed in, so runs can be
attributed to the cluster that served them and queried by it.

The supervisor adds both values to every attempt start it proxies for a
runner, overriding anything the runner sends. The engine merges them
into the run's existing `annotations` inside the update that already
starts the attempt, replacing the previous stamp as a unit. Replication
projects them into two new `task_runs_v2` columns, `worker_cell` and
`worker_lane`. The cell name comes from the new optional
`TRIGGER_WORKER_CELL_NAME` supervisor env var (helm:
`supervisor.config.cellName`); the lane is derived from the selected
backend. Both also go into the worker instance metadata sent on connect,
and admins see them on the run inspector. Neither is exposed to
customers.

Mono-RevId: 87b2c6a5ce7724b4d32c2826fc74c266e3604c3a