Function decode_float_row 

pub fn decode_float_row(
    row: &mut [u8],
    samples: usize,
    bps_bytes: usize,
    factor: usize,
    little_endian: bool,
    out: &mut [f32],
)

Decode one row of DNG floating-point samples straight into out.

row holds the row’s bytes (modified in place). With factor > 0 the row uses the floating-point predictor: bytes are byte-differenced with distance factor, then stored as byte planes (most significant plane first). With factor == 0 the samples are stored contiguously in file byte order (little_endian). Fuses what undo_float_predictor + float_bytes_to_f32 do, without a temporary row.