CON-1516: Remove vCPU gate from self-test preflight

[jjziets]

Summary

• remove the self-test preflight hard gate based on offer["cpu_cores"] >= 2 * num_gpus
• leave physical CPU validation to the self-test runtime image, which can inspect visible physical cores
• remove the >64 direct-port advisory so hosts are only warned when too few direct ports are available
• map 5001/udp when launching the self-test image and externally probe it after TCP /progress is reachable
• select the new CLI/image contract tag family: vastai/test:self-test-cli-1.2.2-cuda-*
• pass VAST_SELF_TEST_CLI_VERSION and VAST_SELF_TEST_CLI_CONTRACT_VERSION into the self-test instance
• make daemon/startup status output conservative by default, with raw daemon evidence printed only under --debugging / --debuging
• redact sensitive debug evidence before it is printed or written into the self-test support bundle
• add manifest fields that explicitly separate CLI/API-visible instance artifacts from local host OS artifacts
• fix false startup-error detection for package names like liberror-perl
• keep raw daemon/status evidence in raw output and the diagnostic bundle without presenting it as a CLI-owned root-cause interpretation

Scope Note

• This PR is the CLI side of physical CPU-core preflight alignment, UDP self-test verification, and the CLI/image contract bump to 1.2.2.
• The paired self-test image/runtime change is in vast-ai/self-test#6. The image opens the UDP responder and rejects launchers that do not provide a compatible self-test contract.
• A live bundle check against vastai/test:self-test-cli-1.2.2-cuda-11.8 confirmed the bundle includes CLI/API-visible instance evidence such as instance/daemon.log, instance/container.log, and instance/show-instance.json.
• The manifest now records those under cli_visible_instance_artifacts / includes_cli_visible_instance_artifacts, while host-local artifacts are recorded separately under local_host_artifacts / includes_local_host_artifacts.
• The bundle does not include host-side OS artifacts (dmesg, journalctl, kaalia logs, or full Docker daemon logs) unless a host-side collection path is used. Exact daemon/Docker startup root-cause mapping still needs daemon-side source/structured error codes or host-side logs.

Validation

• ./.venv/bin/python -m pytest tests/cli/test_self_test_support_bundle.py -q — 9 passed, 1 existing pytest config warning
• ./.venv/bin/python -m pytest tests/cli -q — 321 passed, 1 existing pytest config warning
• git diff --check
• graphify update . — updated graphify-out/graph.json and GRAPH_REPORT.md; skipped HTML because graph has 82,245 nodes
• Live bundle inspection using machine 53944 and vastai/test:self-test-cli-1.2.2-cuda-11.8 confirmed CLI/API-visible log artifacts are present and includes_local_host_artifacts=false

Image Build / Dogfood

• Canonical self-test image build was triggered from vast-ai/self-test#6 with push=true.
• Run: https://github.com/vast-ai/self-test/actions/runs/28854407623
• Inputs: image=vastai/test, tag_prefix=self-test-cli-1.2.2-cuda-, min_cli_version=1.2.2, cuda_version=all.
• Validation jobs passed for CUDA 11.8, 12.8, 13.0, and 13.3.
• As of this PR update, vastai/test:self-test-cli-1.2.2-cuda-11.8, vastai/test:self-test-cli-1.2.2-cuda-13.0, and vastai/test:self-test-cli-1.2.2-cuda-13.3 have built/pushed successfully; CUDA 12.8 is still running in the same workflow.

[robballantyne]

Looks good. I will approve without requesting changes when test image is available

[jjziets]

CON-1531 follow-up pushed in 63f2752.

What changed:

• Broadened self-test progress-stage parsing for the current image output (ResNet18 test on all GPUs, ECC test on all GPUs, NCCL distributed test with N GPUs, stress-ng and gpu-burn tests simultaneously).
• Improved the offline runtime summary so it covers instances going offline before or during runtime.
• Split repeated status polling/API failures into instance_status_poll_failed instead of folding them into startup timeout.
• Cleaned up startup-timeout wording for hosts.
• Surfaced cleanup_failed if the workload passes but the temporary paid instance cannot be destroyed.
• Added tests for current image stage lines, status-poll timeout classification, and cleanup failure.

Validation:

• ./.venv/bin/python -m pytest tests/cli -q -> 313 passed, 1 existing config warning.
• Paid dogfood with latest PR image vastai/test:self-test-udp-dogfood-cuda-12.8:
  • Green machine 35008: passed TCP, UDP, system requirements, ResNet18, ECC, NCCL, stress/gpu-burn; cleanup succeeded.
  • Red machine 141071: preflight caught low reliability/upload; forced runtime failed with progress_endpoint_unreachable; support bundle printed and cleanup succeeded.

Local evidence/artifacts:

• /Users/hanneszietsman/VastAi/CON-1531-self-test-error-flow-map.md
• /Users/hanneszietsman/VastAi/dogfood-captures/host-self-test-latest-pr/20260625T090214Z.tar.gz

[jjziets]

Final CON-1531 diagnostic follow-up SHA is now f737010 after amending in the catch-all unexpected_error path.

Additional tightening in the final amend:

• Added first-class unexpected_error catalog copy and runtime diagnostic rendering for otherwise unhandled CLI exceptions.
• Added preflight_checks stage marking before threshold checks run.
• Added regression coverage for unexpected exception shaping/redaction.

Latest local validation:

• ./.venv/bin/python -m pytest tests/cli/test_runtime_diagnostics.py tests/cli/test_self_test_support_bundle.py tests/cli/test_machines_commands.py -q -> 84 passed, 1 existing config warning.
• ./.venv/bin/python -m pytest tests/cli -q -> 314 passed, 1 existing config warning.