GeezerVison

`wait_for_completion` returned as soon as `completion_status` named the
step that took a request's input, before that step finished.  A test that
then stopped the controller could abandon the step before it was
journaled, so the restarted pipeline never saw the input.  This is how
`replay_does_not_feed_http_input_during_commit` lost its last record; two
other replay tests stop the controller the same way.

The helper now also waits for `total_completed_steps` to reach that step,
the rule that `/completion_status` applies.

Fixes #7631

Signed-off-by: Leonid Ryzhyk <[email protected]>

The static engine runs circuits in persistent mode, where checkpointing
an operator without a persistent id fails with `NoPersistentId`, and the
failed checkpoint terminates the circuit.  The views and the aggregate of
`TestCircuit::Aggregate` had none, so any checkpoint of that circuit,
such as a periodic one in a test that outlives the checkpoint interval,
killed the pipeline.

Signed-off-by: Leonid Ryzhyk <[email protected]>

…tent ids.

The static engine runs test circuits in persistent mode, where checkpointing
an operator without a persistent id fails, and the failed checkpoint
terminates the circuit.  `TestCircuit::IndexedView` gave its table, view, and
index no persistent ids, so it could not be checkpointed at all, and three
fault-tolerant controller tests used a view without one, which a periodic
checkpoint would have hit.

Signed-off-by: Leonid Ryzhyk <[email protected]>