feat: add DUCKDB dialect and bidirectional DuckDB converter

[michellepellon]

Add DUCKDB dialect and a bidirectional DuckDB converter

What

This PR adds DuckDB as a supported expression dialect and contributes a reference converter for it, in three parts:

1. Dialect registration — adds DUCKDB to the Dialect enum in core-spec/osi-schema.json, core-spec/spec.yaml, the dialect table in core-spec/spec.md, and the OSIDialect / OSIVendor enums in python/src/ossie/models.py.
2. Validation support — maps DUCKDB to sqlglot's duckdb dialect in validation/validate.py, so DuckDB expressions get full syntax validation (sqlglot's DuckDB support is first-class; no skip-list entry needed).
3. Converter — converters/duckdb/, a bidirectional Ossie ↔ DuckDB converter following the hub-and-spoke pattern in converters/README.md:
   • Export: Ossie model → DuckDB SQL (one view per dataset with field expressions, COMMENT ON from descriptions, one view per metric with joins derived from declared relationships). Dialect selection prefers DUCKDB, falls back to ANSI_SQL with a warning.
   • Import: DuckDB database → Ossie YAML via information_schema and duckdb_constraints() (primary keys, unique keys, and foreign keys → relationships; table/column comments → descriptions).
   • CLI (ossie-duckdb export|import), pytest suite with round-trip tests executing generated SQL against in-memory DuckDB, and schema validation of imported models against osi-schema.json.

Why

DuckDB (and MotherDuck, its managed cloud service) is widely used as an analytics engine in lakehouse and embedded-analytics stacks, and is a common substrate for the AI-agent workloads that ai_context targets. Because DuckDB connections are passed through, the converter works against MotherDuck databases (md: connection strings) unchanged.

Notes

• No changes to existing converters; unknown-dialect handling elsewhere is unaffected (existing fallback chains ignore dialects they don't target).
• The example models in examples/ are untouched and still validate.
• Aware of Only allow one dialect per OSI document #52 (single dialect per document); the enum addition is orthogonal — whichever way that lands, DuckDB needs to be a valid dialect value.