GRIB2 encodes binary (E) and decimal (D) scale factors in sign-and-magnitude
format (MSB = sign bit, lower 15 bits = magnitude), not two's complement.
The previous read_i16 call misinterpreted 0x8009 as -32759 instead of -9,
causing 2^E to underflow to 0.0 in f32 and zeroing all decoded values.
Fix: replace read_i16 with read_grib_scale for E and D in
decode_simple_packing. Add regression tests covering sign-magnitude decoding
and a realistic DWD CLCT packing scenario.
