fix(crn): skip malformed CRN entries in CrnList.from_api

[aliel]

A malformed CRN payload (e.g. missing GPU.model) previously raised
ValidationError and crashed any consumer iterating the list, including
aleph instance list. A hostile CRN operator could exploit this to
break clients across the network.

Validate each entry independently, log a warning identifying the
offender, and skip non-dict entries. Add tests covering malformed,
non-dict, and empty payloads.

[github-actions]

Failed to retrieve llama text: POST 502: Bad Gateway

[foxpatch-aleph]

This PR hardens CrnList.from_api against malformed CRN payloads by wrapping model_validate in a try/except, logging a warning with the offending entry's identifier, and continuing. The fix is correct, minimal, and well-tested. No security issues, no logic errors. Ready to merge.

src/aleph/sdk/client/services/crn.py (line 136): Nit: item.get("hash") treats an empty string hash as falsy and falls through to name. Consider item.get("hash") is not None if an empty string hash should be treated as a valid identifier. No practical impact — just a consistency observation.

[foxpatch-aleph]

Correctly fixes a real availability issue where a single malformed CRN entry (e.g., missing GPU.model) would crash the entire CrnList.from_api() consumer. The try/except approach with per-entry validation, warning logging, and identifier extraction is appropriate. Tests cover malformed dict entries, non-dict items, None entries, and empty payloads. The code is clean and the fix is minimal — no over-engineering.

src/aleph/sdk/client/services/crn.py (line 11): Minor style nit: logger definition sits between imports (between line 9 and line 13). The existing code already has this pattern, so it's consistent with the codebase, but worth noting.

[foxpatch-aleph]

A clean, well-targeted fix for a real availability concern: malformed CRN entries returned by the CRN list endpoint no longer crash consumers like aleph instance list. Each entry is validated independently in a loop, with ValidationErrors caught and logged with a best-effort identifier. The test suite covers malformed entries, non-dict items, None values, and empty payloads. No correctness, security, or code quality issues.