fix: Robin backend join materializer and plan path support (#473)

[eddiethedean]

Summary

Fixes #473: join no longer raises SparkUnsupportedOperationError: Operation 'Operations: join' is not supported when using the Robin backend with same-name join keys.

Changes

• lazy.py: When backend is Robin and the operations queue contains join or union, skip the Robin plan path and run the materializer (op path) directly. This avoids to_robin_plan raising for join and ensures join is handled only by the materializer, which already supports same-name on.
• materializer.py: Document in _join_on_to_column_names that different-name join keys (e.g. left["id"] == right["id_right"]) are not supported; same-name on or on="col" / on=["c1", "c2"] are supported.
• test_robin_materializer.py: Add test_join_same_name_on_robin that runs left.join(right, on="id", how="inner").collect() with Robin and asserts row count and contents.
• v4_behavior_changes_and_known_differences.md: Document that join uses the op path only, same-name keys supported, different-name keys not supported (with link to Robin backend: Join materializer and plan path support #473).

Expected behavior (per issue)

• left.join(right, on="id", how="inner") runs on the op path and returns joined rows when backend is Robin.
• Same-name join keys work (materializer already supported this).
• Different-name keys are documented as a limitation; tests using them can be skipped or adapted.

Made with Cursor