Motivation

• Replace fragile per-row composite key construction with explicit DataFrame key columns to simplify and speed up propagation logic.
• Avoid ad-hoc stringification code (value_to_key/composite_key) and rely on polars expressions to produce stable string keys.
• Simplify the propagate_target_rows API to operate on named key columns rather than slices of column names.

Description

• Remove value_to_key and composite_key and add key_from_row that reads a precomputed string key from a row by column name.
• In propagate_target_from_ancestor build temporary key columns __self_key and __parent_key using a local build_key_expr which uses polars.col(...).cast(String).fill_null("<null>") and polars.concat_str(separator="##"), then call with_columns to attach them to the DataFrame.
• Change propagate_target_rows signature to accept self_key_col and parent_key_col and update its internal lookups to use key_from_row.
• Drop the temporary key columns before returning the result DataFrame and adapt unit tests to populate and use __self_key/__parent_key via a helper rows_with_keys.

Testing

• Ran the crate unit tests with cargo test which executed the propagation tests.
• The test functions blocker_terminates_chain_resolution, node_level_self_blocker_prevents_propagation, node_level_self_blocker_not_later_does_not_block, node_level_parent_blocker_prevents_propagation, and blocker_checks_use_start_target_as_anchor_across_ancestors were executed and passed.

Codex Task