Modified the way data is accessed from `pivoted_plot_data` to avoid problems with NaN in some lines

[MarinManuel]

This is a proposed fix for #103

The issue seems that the pivoted_plot_data table contains columns for all idx pairs, but some id (rows) could only have data in some of the idx pairs and not others.

I have changed the way this data is accessed when looping through the idx pairs, and drop NaN rows to get rid of unnecessary rows.

It seems pytest passed all the tests, so hopefully this change does not break anything. Maybe this dataset (or a synthetic equivalent) could be added to the test suite?