The specification does not prevent an encoder to write the amplitude 0
as 0 amplitude_bits.

Our get_bits() implementation might not support a zero sized read
properly, thus the additional branch.