Keep converter outputs aligned with root schema

[LiangRuoShui]

Why

The published core JSON Schema only allows version and semantic_model at the document root and sets additionalProperties: false. dbt and Honeydew could still emit root-level advertisement fields (dialects / vendors), and the Python OSIDocument model accepted those fields, which allowed generated OSI documents to fail the official validator.

What changed

• Removed root-level dialects emission from the dbt MSI-to-OSI converter.
• Removed root-level vendors emission from the Honeydew-to-OSI converter.
• Made Honeydew reject root-level advertisement fields such as vendors and dialects on OSI input.
• Made the Python OSIDocument model reject schema-invalid root extra fields.
• Updated dbt snapshots to reflect schema-valid root output.

Tests added/updated

• Added dbt regression coverage asserting serialized OSI output omits root dialects and vendors while preserving expression-level dialects.
• Added Honeydew regression coverage asserting generated OSI output omits root advertisement fields.
• Added Honeydew input validation coverage asserting root vendors / dialects are rejected.
• Added Python model coverage asserting root dialects is rejected.

Verification

• uv run validation/validate.py examples/tpcds_semantic_model.yaml
• cd python && uv run --with pytest pytest -q
• cd converters/dbt && uv run pytest -q
• cd converters/honeydew && uv run pytest -q

[LiangRuoShui]

Gentle ping on #306, #307, and #308.

I noticed the workflows for #306 and #308 are waiting for maintainer approval (action_required). Could someone approve the CI runs when convenient?

For review order, #307 and #308 are related:

• Validate relationship column pair counts #307 adds validator coverage for mismatched relationship column counts.
• Reject mismatched relationship columns in converters #308 adds converter-level fail-fast checks for the same malformed input.

#306 is independent and keeps converter/Python output aligned with the published root schema.

Happy to adjust or split further if the direction looks off.

[khush-bhatia]

Hey @LiangRuoShui Could you please resolve the merge conflicts because of the rename ?

[LiangRuoShui]

Thanks for the reminder! I have rebased this PR on the latest main and resolved the merge conflicts caused by the rename. I also ran the relevant local tests to make sure the changes still behave as expected. Please let me know if there is anything else you would like me to adjust.

[khush-bhatia]

The header shoudln't have been removed ?

[kayemkim]

Since this was opened, #297 added root-level dialects/vendors to the schema, so the two sides of this PR's premise now disagree on main. I have written the details up in #393 rather than here, since it also touches #148, the Sigma converter and the orionbelt tests.

[kayemkim]

#397 has taken the two root properties out of the schema and out of the Sigma converter, so the premise here holds on main again. One data point for the rebase: on main as it was before #397, the extra="forbid" on OssieDocument in this PR failed 32 sigma tests, because that converter still set both fields. With current main merged into this branch, python, dbt, honeydew, sigma and validation all pass together.

[LiangRuoShui]

Closing this PR for now. Thank you for the time and review.