Consider indicating the size of floats in the API in the cb->length field

[mjkoster]

    case AI_2:
#ifndef CBOR_NO_FLOAT
      cb->type = CN_CBOR_DOUBLE;
      cb->v.dbl = decode_half(val);
      cb->length = 2; // Tiny Float, especially nice for serializing binary fractions
#else /*  CBOR_NO_FLOAT */
      CN_CBOR_FAIL(CN_CBOR_ERR_FLOAT_NOT_SUPPORTED);
#endif /*  CBOR_NO_FLOAT */
      break;
    case AI_4:
#ifndef CBOR_NO_FLOAT
      cb->type = CN_CBOR_DOUBLE;
      u32.u = val;
      cb->v.dbl = u32.f;
      cb->length = 4; // Short Float, good enough for most constrained network purposes
#else /*  CBOR_NO_FLOAT */
      CN_CBOR_FAIL(CN_CBOR_ERR_FLOAT_NOT_SUPPORTED);
#endif /*  CBOR_NO_FLOAT */
      break;
    case AI_8:
#ifndef CBOR_NO_FLOAT
      cb->type = CN_CBOR_DOUBLE;
      u64.u = val;
      cb->v.dbl = u64.d;
      cb->length = 8; // Long Float, no excuses
#else /*  CBOR_NO_FLOAT */
      CN_CBOR_FAIL(CN_CBOR_ERR_FLOAT_NOT_SUPPORTED);
#endif /*  CBOR_NO_FLOAT */
      break;
    default: cb->v.uint = val;