natural-earth-vector | public domain map dataset available at three scales | Map library

I have a Android App that uses GPS and reports it to the backend. For some reason (mock location or low gps accuracy I guess) a lot of measurements were saved with coordinates beyond lands.

I'm writing a python pandas/geopandas short code to filter out those but it seems not to be that trivial.

My initial idea was to join registered GPS points with hi-res (10m) land shapes.

I recently made some maps using D3 and had some problems with paths not closing properly as you can see in the first example below. I assume this is some kind of data issue around precision at -90 degrees latitude, but I'm not sure how to validate that suspicion. Interestingly, when I include the most recent version of the d3-geo module on the page, the map renders as expected (see second example below). What's going on here?

I have a shape that boundaries goes from -180º to 180º on longitude, but on latitude, It goes from -90º to 83.64513º. In a front-end application using leaflet, when It asks for mapnik server the image of the tiles, I convert tile positions to latitude/longitude. Longitude works, but latitude doesn't. I'm using this formula to convert: