matten: Heterogeneous data with `--features dynamic`

The third part of the matten series introduces the dynamic feature for handling heterogeneous real-world data. While standard numeric Tensors require clean f64 cells, the dynamic feature provides an ingestion and cleanup layer for data from JSON APIs or CSV files containing missing cells, integers, floats, or boolean flags. Users can ingest mixed data using from_json_dynamic and from_csv_dynamic, inspect and clean the Element variants, and then explicitly convert the results into numeric Tensors for arithmetic operations.