Migrate converters and Python SDK to flat semantic model documents

[flyrain]

Summary

Migrate the Python SDK and all affected converters to the single-model document format merged in #383. Ossie input/output now puts name, datasets, relationships, and metrics directly at the root alongside version; the legacy semantic_model wrapper is rejected instead of selecting its first entry.

version: 0.2.0.dev0
name: sales_analytics
datasets:
  - name: orders
    source: sales.public.orders
relationships: []
metrics: []

Rebased onto current main after #383 merged. This PR now contains only the SDK/converter migration and its CI/documentation updates. It also incorporates #397: root-level dialects and vendors are removed; expression dialects and vendor custom extensions remain supported.

Behavior

• Flatten the shared OssieDocument API and update dbt, Sigma, Wisdom, Databricks, Honeydew, Omni, OrionBelt, Snowflake, GoodData, NVIDIA GSF, Microsoft/Power BI, Salesforce, and Polaris readers/writers, fixtures, snapshots, and usage documentation.
• Remove obsolete root dialects/vendors fields and first-model/drop-extra-model behavior. Preserve vendor-native collections such as dbt's semantic_models.
• Preserve the ontology mapping's embedded semantic_model structure and exclude standalone document metadata from that embedded object.
• Add Polaris import --output-dir DIR to emit one document per nonempty namespace with deterministic, collision-safe filenames and no overwrites. Single-file/stdout imports require one nonempty namespace.
• Trigger converter CI, including Microsoft, on shared schema/example changes and run shared SDK tests in Validation CI.

This is a breaking SDK and converter format change for the mutable 0.2.0.dev0 specification. Legacy documents must be migrated first; see the spec's migration guidance.

Checklist

• Converter logic, fixtures, examples, and documentation reflect Define one semantic model per document without a wrapper #383.
• Breaking behavior and bulk export migration described.
• Focused regression coverage added or updated.
• Ontology embedding and vendor-specific model collections preserved.
• No runtime dependencies added.

Validation

• Passed 199 focused SDK/converter tests, including dbt relationship, composite-key, template-safety, CLI, and snapshot regressions.
• Passed Microsoft CLI import/export round-trip and schema validation; all 10 checked-in standalone converter fixtures validate against the merged schema.
• Microsoft Ruff checks, changed-Python compilation, and whitespace checks pass. Comparison against the rebased branch found no new Ruff diagnostics in the other changed Python files; existing style findings remain.
• Full converter regression suites and optional Microsoft TOM/live-engine checks were not run locally; full regression validation is left to CI.

[kayemkim]

#393 was settled by removing the root dialects/vendors arrays from the schema (#397, now on main), and #383 has been asked to drop them in its rebase. This branch still keeps them in a few places that will need the same treatment: converters/sigma/tests/test_roundtrip.py asserts serialized["dialects"] == ["ANSI_SQL", "SIGMA"] and serialized["vendors"] == ["SIGMA"], the orionbelt test_schema_rejects_root_dialects_and_vendors is flipped to test_schema_accepts_root_dialects_and_vendors, the nvidia reader allow-lists both keys at the root, the honeydew tests expect a root vendors, and the SDK tests and python/README.md list them alongside version. I merged the schema change with this branch earlier; the conflict is confined to those spots and resolves by dropping them.