Write and read polar tables as Arrow when table_format is :arrow #371

Write and read polar tables as Arrow when table_format is :arrow
Today table_format (#257) only covers the per-node Cp/cf tables; every polar still goes out as CSV through write_polar_csv / write_polar_matrix_csv (src/airfoil_aero/airfoil_solvers/common.jl, src/airfoil_aero/airfoil_io.jl) and comes back through the line-based load_polar_data (src/yaml_geometry.jl:77). BeyondTheSim runs SK100 with table_format: arrow and still gets CSV polars.

Make the polars follow table_format by reusing the node-table I/O, not by adding a second Arrow path:

• read_node_table / write_node_rows (src/section_aero.jl:151, :186) already pick CSV or Arrow from the file suffix and take exactly the (alpha, delta, values) row form a POLAR_MATRICES table is, with values = [Cl Cd Cm]. Extend them (e.g. named value columns, so a polar keeps its Cl/Cd/Cm header) rather than writing polar-specific Arrow code; the polar writers and load_polar_data go through them.
• convert_node_table / migrate_node_tables (src/obj_adapter/obj_to_yaml.jl:115) should migrate a cached dataset's polars too, so switching format rewrites them instead of re-running NeuralFoil.
• obj_to_yaml / surfplan_to_aero_yaml write the polar path with the matching suffix into the geometry YAML; existing CSV polar datasets keep loading untouched.
• Cover POLAR_VECTORS as well, or say on the card why not.

Builds on #369 (the 16-significant-digit csv_fields row writer on the same functions): agent stack 369 before the first push, and fold csv_fields into the shared CSV branch rather than keeping two row formatters.