fix(hack): let teardown delete crashed actors

[NekoPunch (orangeCatDeveloper)]

Fixes #794

A crashed actor makes --delete-all abort mid-teardown, leaving the ActorTemplate behind; the next --deploy-demo-counter is then rejected on the immutable spec and the cluster keeps serving the stale golden snapshot.

prepare_actor_for_delete treats STATUS_CRASHED as unexpected and returns 1, which errexit turns into a hard abort — even though DeleteActor accepts CRASHED (workflow_delete.go). Teardown now treats crashed actors as deletable as-is.

The demo flag dispatch had the mirror-image defect: handlers ran from an if condition, which suppresses errexit for their whole call tree, so this same teardown — and failed deploys like the issue's immutable-spec apply — exited 0. Handlers now run as plain commands and report an unclaimed flag through ate_demo_flag_unhandled.

Before (main, actor in STATUS_CRASHED):

[step]:   preparing actor ate-demo-counter/repro-794 for delete
cannot delete actor repro-794: unexpected status STATUS_CRASHED
# exit 1; ActorTemplate "counter" survives, ate-system untouched

After (this branch, same state):

[step]:   preparing actor ate-demo-counter/repro-794 for delete
actor "repro-794" deleted
...
actortemplate.ate.dev "counter" deleted from ate-demo-counter namespace
# full teardown completes, exit 0

Repro: deploy the counter demo, resume an actor, delete its worker pod (~20s later the syncer marks it STATUS_CRASHED), run --delete-all.