feat: treat character arrays as strings

[Michael-F-Bryan]

ULog char[N] fields are currently flattened as one numeric column per byte, which loses the schema's distinction between text and arbitrary uint8_t[N] data. In the Zoda log, values such as com.vimdrones.esc_s50#M3 therefore become 24 separate CSV columns.

This keeps scalar char byte-valued but treats char[N] as one string across dynamic records, datasets, CSV, Arrow, and Parquet. Trailing NUL padding is removed, embedded NULs are preserved, and malformed UTF-8 is rejected only at the Arrow boundary where a UTF-8 string is required. Typed Go adapters declare the fixed wire width explicitly:

Name string `ulog:"name,char[80]"`

This deliberately changes the pre-release dataset schema: character arrays now produce name rather than name[0] through name[N-1]; ordinary numeric arrays retain their existing flattened shape.

The captured Zoda log remains outside the repository. It still produces 99 datasets and 3,598 records, with all 99 datasets readable through CSV, Arrow, and Parquet.