Float16_t with up to 32 bits is allowed (not just 16).

[jpivarski]

Tested on files made by https://root.cern/doc/master/float16_8C_source.html with

   Float16_t fI14; //[-pi,pi,14] saved as a 14 bit unsigned int

replacing 14 with 20, 32, 33, 0, 1, and 2. Values outside of the legal range, 2 <= num_bits <= 32, are left in the file and have to be mapped to 32, even for Float16_t.