fix (visual.ts): avoid adding overlapping categories from data view

[dm-p]

In the event that a categorical column is used in multiple data roles, this ensures that when we enumerate source and destination properties, any duplicates are discarded and only one occurrence is eligible for the node file.

Should ideally resolve #85 (or get us closer to root cause if not)

[lmeyerov]

@dm-p can you confirm, at least in the original #85 case, that the expected results now come out?

[dm-p]

I could not reproduce #85 for the same data assignments as you, including switching-up the order of categories. This PR resolves what was discovered through investigation and is what I believe is causing the values to be swapped around (as the order in which duplicate fields would get added would have caused an overlap).

Alternatively, if you can send me your .pbix with the reproduced issue, I can test the developer visual against this specific occurence and confirm.

This change has been tested to confirm that only one instance of a category's values from the data view will be passes through to the node file per field if this situation is replicated (and again could not reproduce #85 on this setup).

[lmeyerov]

Gotcha, I will try

The suspicious thing to me is because of the switchup, it sounds like somewhere we're assuming an indexing order to get a column, vs doing an appropriate lookup. So it's actually an interesting bug where even if we eliminate what we consider a bad user input, the situation is revealing a risky indexing operation. So maybe there is a legit scenario where it'd trigger too, and we've just been lucky.

[lmeyerov]

FYI, a fun way to viz these may be via https://github.com/graphistry/pygraphistry#group-in-a-box-layout

using something like partition_key='top_motif_id'