fix(resources): preserve multipart file uploads in generated client (HYBIM-882)

[etserend]

Summary

Port of upstream commit 02ce8bb — fix: Preserve multipart file uploads in generated client (#624).

• openapi.yaml: Replace contentMediaType: application/octet-stream → format: binary for 7 Body_* multipart file fields, enabling openapi-python-client to generate File-typed attributes instead of plain str
• scripts/import-openapi-yaml.sh: Add yq patches for the 7 binary fields so the fix survives future openapi.yaml regenerations; tighten error handling ($? -ne 0)
• 6 generated Body_* models: file: str → file: File; add BytesIO/File/FileTypes imports; call .to_tuple() in to_dict/to_multipart; construct File(payload=BytesIO(...)) in from_dict
• tests/test_datasets.py: 2 new tests verifying File payloads round-trip through to_multipart() correctly

Test plan

• poetry run pytest tests/test_datasets.py -v — 61 passed
• New tests: test_create_dataset_body_serializes_file_as_multipart_upload, test_code_scorer_bodies_serialize_files_as_multipart_uploads
• Trigger regenerate-api-client workflow after merge to produce clean generated output

[fercor-cisco]

🤖 This review was generated by the Astra agent (claude-opus-4-8). It may contain mistakes.

Verdict: approve — Faithful, correct port of upstream generated multipart-upload fix; call sites already pass File objects, so this repairs genuinely broken uploads. Only minor testing/docs gaps.

[fercor-cisco]

Port faithfulness review vs upstream 02ce8bb (rungalileo/galileo-python#624)

Verdict: Faithful port ✅

Every substantive change from upstream #624 is present. Remaining deltas are cosmetic artifacts of a different openapi-python-client generator version, not behavioral differences.

File-by-file comparison

Area	Upstream #624	This branch	Match
openapi.yaml — 7 fields contentMediaType → format: binary across 6 schemas	✓	✓ (identical)	✅
scripts/import-openapi-yaml.sh — yq patches + verification block + $? -ne 0	✓	✓	✅
6 Body_* models — file: str→File, .to_tuple(), File(payload=BytesIO(...))	✓	✓ (logic identical)	✅
tests/test_datasets.py — 2 new tests	✓	✓ (path adapted galileo→splunk_ao)	✅

Both new tests pass locally (2 passed).

Cosmetic-only differences (not defects)

Behavior-preserving, stemming from a different generated baseline in this repo:

1. Union ordering / docstrings — local uses None | str | Unset and Attributes:, upstream has None | Unset | str and NumPy-style Attributes\n----------. Pure formatting.
2. body_manual_llm_validate_multipart... — upstream imports/uses FileTypes for a to_dict local var annotation; local uses list[Any] and omits the FileTypes import. Annotation only — no runtime difference, no NameError.
3. body_validate_code_scorer_dataset... / ..._log_record... — local diffs contain an extra incidental refactor of the unrelated _parse_required_scorers / _parse_scoreable_node_types helpers (x = cast(...); return x → return cast(...)). Regeneration noise, not in #624, but behavior-identical.

Items outside #624 scope (expected, harmless)

• scripts/import-openapi-yaml.sh also adds ListAnnotationQueueParams.properties.sort.default = "None" (from commit 5f89e46) — a separately-needed local patch, correctly bundled into the monolithic yq command.
• tests/test_decorator_distributed.py relaxes a duration assertion (> 0 → >= 0, commit b2f1555) — an unrelated CI-stability fix that rides along on the branch.

Conclusion: Complete and correct port. No correctness gaps. The only deltas versus upstream are generator formatting artifacts and two clearly-separate commits that legitimately ride along on the branch.

🤖 Generated with Claude Code