add flatten-dataframe

[slavlotski]

Closes #47.

Adds a new public helper chispa.flatten_dataframe(df, sep="_") that turns a DataFrame containing nested columns into a fully flat one. It walks the schema and, for each complex
field:

• StructType → expands every sub-field into its own top-level column named (e.g. name.firstname → name_firstname).
• ArrayType → applies explode_outer, so each element becomes its own row (null-preserving).
• MapType → collects the distinct keys present in the column and turns each into a column.

The walk repeats until no nested types remain, so deeply nested / mixed schemas (struct of map of array of struct, etc.) are flattened in one call.