Document precision losses in metadata/etc. for standalone Qurro

[fedarko]

This likely impacts other stuff besides the metadata that gets "sent" from python to JS.

As an example of this "issue," consider feature fe98555e63f547c33b3987391cd34d5d in the Q2 moving pictures data. It has a Confidence of 0.9993999015917799, but here's the weird thing: in the Q2 Qurro visualization of the moving pictures data, this confidence value shows up correctly in the DataTable when this feature is selected. In the standalone Qurro visualization of this same data, this instead shows up as 0.99939990159178 (rounded off at the end).

If you count it out, this rounded-off value is exactly 16 characters long, which seems like it's probably relevant to this problem (maybe this is a pd.read_csv() thing? It should be reading everything in as strings at first, but you never know).

This is such a minor issue that I doubt it will impact anyone, but it would be good to address this problem -- and/or document why it's happening. Some imprecision is to be expected (see: how JS breaks with > ~2^53 values or something), but it's good to quantify this wherever possible.