Allow CLI load runs to continue after terminal iteration failures

[THardy98]

What was changed

omes run-scenario and run-scenario-with-worker now accept --iteration-failure-policy=fail-fast|continue. Existing commands remain fail-fast by default; operators running sustained load can explicitly select continue. Retry count and terminal-failure policy remain independent.

The loadgen library still defaults to fail-fast. When continuation is enabled, it returns a typed IterationFailuresError containing attempted, succeeded, failed, and elapsed values. The CLI treats only that degraded completion as successful in continue mode, so cancellation, timeout, validation, connection, joined, and other run-level errors remain fatal. Every terminal failure remains an error log, and a completed degraded run emits a structured warning with its outcome counts.

The operational observability changes are intentionally separated into stacked follow-up PR #486.

Why?

A terminal iteration failure currently ends an omes run-scenario invocation. In a Kubernetes load-driver deployment, that turns a transient workflow failure into a container restart and stops the configured load run early. This PR adds an explicit continuation mode for that use case while preserving existing fail-fast behavior for unflagged CLI, CI, and automation callers.

Checklist

1. Closes [Feature Request] Gracefully handle load driver failures #453

2. How was this tested:

   • Focused loadgen tests cover fixed-iteration and duration-based degraded results.
   • CLI tests cover the zero-value and registered fail-fast defaults, explicit continue, explicit fail-fast, invalid policy, wrapped degraded result, joined error, and unrelated error cases.
   • A logger-observer test verifies the final degraded-run warning and its attempted, succeeded, and failed fields.
   • Full cmd/omes tests and focused race-detector coverage pass locally.
   • All 46 GitHub Actions checks pass on the revised head.

3. Any docs updates needed?

   • Updated docs/running.md and docs/authoring-scenarios.md for the opt-in policy, warning behavior, and relationship to retries.