Push logs_url to status API eagerly and via final PUT

[ishandhanani]

Summary

• Fix: recipe-defined reporting.status.endpoints were silently skipped by the post-process dashboard PUT. The path only read load_cluster_config(), so any endpoint set in the recipe (rather than cluster srtslurm.yaml) never received logs_url. Consequence: the collector recorded job lifecycle correctly but logs_url stayed null, and consumers could not find the S3 artifacts.
• Fix: report_completed fired before run_postprocess in do_sweep.run()'s finally block, so the completed PUT reached the collector before any rollup was generated.
• Feature: new StatusReporter.report_artifacts(logs_url) for eager mid-run pushes — fires as soon as S3 sync completes, before AI analysis runs, so a hanging or crashing analyzer cannot strand the artifact pointer.

Changes

File	What
src/srtctl/core/status.py	report_completed(exit_code, logs_url=None) kwarg; new report_artifacts(logs_url) that PUTs at status=benchmark (does not flip lifecycle)
src/srtctl/cli/mixins/postprocess_stage.py	run_postprocess(exit_code, reporter=None); eager reporter.report_artifacts after S3 sync; stash self._last_logs_url; removed the broken _report_metrics
src/srtctl/cli/do_sweep.py	finally reordered so run_postprocess runs before report_completed; passes reporter in and pulls logs_url back for the final PUT
tests/test_status.py	Coverage for logs_url kwarg and report_artifacts (disabled, empty URL, success without flipping to completed)
tests/test_postprocess.py	Replaced _report_metrics assertions with the eager-push + stash shape; added skip-paths for no-S3-URL and no-reporter cases
src/srtctl/core/config.py, src/srtctl/core/__init__.py	Small helper refactor pending in the tree: extract find_cluster_config_path, add get_cluster_aliases

Why two PUTs

Defense in depth: the eager report_artifacts PUT fires the moment S3 sync returns, independent of whether the AI analyzer runs, times out, or crashes. The final report_completed in do_sweep then reasserts the same logs_url idempotently at status=completed. If either PUT succeeds, the collector has the pointer.

Compat

• report_completed(exit_code) still works — logs_url is optional. No call-site changes required for existing callers.
• No changes to the Status API v1 wire contract. logs_url has always been a first-class field in JobUpdatePayload.

Test plan

• uv run pytest tests/ -q — 589 passed, 2 skipped, 6 deselected
• uv run ruff check on touched files — all pass
• uv run ruff format --check on touched files — all pass
• Live verification against the Brev collector once a benchmark recipe with reporting.status.endpoint lands on a cluster with reporting.s3 configured — expected: GET /api/jobs/{id} returns logs_url pointing at the S3 prefix.

🤖 Generated with Claude Code