Fix map.unproject regression in 2D mode

[mourner]

An attempt at fixing #10215, using @karimnaaji's suggestion. I'm not sure why the sourceCache.tilesIn tests behave differently and if it's an issue in practice though. Also @arindam1993 you had an alternative suggestion for a fix in mind — would do a PR for it?

Launch Checklist

• briefly describe the changes in this PR
• write tests for all new functionality
• manually test the debug page
• tagged @astojilj if this PR includes shader changes or needs a native port
• apply changelog label ('bug', 'feature', 'docs', etc) or use the label 'skip changelog'
• add an entry inside this element for inclusion in the mapbox-gl-js changelog: <changelog>Fix a regression with map.unproject and map.panBy acting unpredictably in certain cases.</changelog>

[karimnaaji]

Looks good to me, can we get unit tests covering the initial examples you hit with this issue? I'm thinking of this specific case:

map.unproject([0, -1000]);
map.unproject([0, 0]);

I'll also give some time for @arindam1993 to add more info concerning some potential failure case he was suggesting on chat prior to merging.

[arindam1993]

Patch for alternative approach is in
#10224

It addresses this case PR wherenin if the horizon is just above 0 and a panBy reverses the direction of motion:

[karimnaaji]

Closing in favor of #10224.