Add when_matched support to non-idempotent time-range materialization

[0xRobin]

Summary

• Add optional materialization_properties.when_matched support to non_idempotent_incremental_by_time_range.
• Parse and normalize when_matched using SQLMesh's expression validation and merge-alias replacement semantics, then pass it through to adapter.merge(...).
• Add unit coverage to verify custom WHEN MATCHED clauses are respected in generated MERGE SQL.

Why

non_idempotent_incremental_by_time_range currently always updates all columns on key match. This makes it impossible to keep immutable ingestion metadata such as _inserted_at unchanged on updates. Supporting when_matched gives model authors explicit control over update behavior while preserving existing defaults when unset.

Implementation Notes

This follows SQLMesh's existing INCREMENTAL_BY_UNIQUE_KEY flow:

• IncrementalByUniqueKeyKind.when_matched parsing/validation:
  • https://github.com/TobikoData/sqlmesh/blob/main/sqlmesh/core/model/kind.py#L510-L560
• Passing when_matched into merge in strategy:
  • https://github.com/TobikoData/sqlmesh/blob/main/sqlmesh/core/snapshot/evaluator.py#L2297-L2330
• Engine adapter merge default-vs-custom when_matched handling:
  • https://github.com/TobikoData/sqlmesh/blob/main/sqlmesh/core/engine_adapter/base.py#L2205-L2273

Validation

• uv run pytest tests/materializations/test_non_idempotent_incremental_by_time_range.py -q