Implement null for normalized_reads

From #3925 :

Should we have null instead of 0 to denote that fact that there is no normalization for this sample?

Yes, null seems to be a very good option.

"normalized_reads": [
        null,
        107.91,
        0
      ],

Here the first sample has no normalization information, and the value of reads should be taken into account. But the third sample has a 0 value.

cc @mikael-s @flothoni @meidanis