Summary

• add submit_and_wait_prepared_body and the shorter submit_and_wait alias for programmatic callers
• validate wait parameters before creating a run and leave the caller's prepared body unchanged; logger= applies only when the helper builds the manager — a supplied manager= keeps its own logger
• support max_wait=None for an unbounded wait
• pass submit_recovery_attempts through to failed-submit recovery so a registered run can be adopted instead of submitted twice
• fix a pre-existing KeyError in PipelineRunManager.submit_prepared_body on locator-style bodies (componentRef with name/digest and no inline spec); such bodies now submit verbatim with a spec-less run context
• tighten wait_for_completion: non-finite max_wait/poll_interval (NaN/inf) now raise PipelineRunError where inf previously acted as a never-firing deadline; max_wait=None remains the unbounded spelling
• wait-validation errors now name the parameter alongside the CLI flag (e.g. max_wait (--max-wait) must be non-negative) since the CLI and programmatic API share the validation path

Context

Scripts currently have to reproduce the CLI's submit, recovery and wait orchestration. A single SDK entry point makes that flow reusable without weakening the existing submission-id recovery guarantees.

Testing

• uv run pytest tests/test_submit_and_wait_prepared_body.py tests/test_pipeline_runs_cli.py
• uvx ruff check on the touched source files
• uv lock --check
• git diff --check