Re-implement "avoid edges" for MapMode::Tile

[ChrisLoer]

Fixes issue #12461.

• Only implement "avoid edges" in MapMode::Tile since it's no longer relevant in Static or Continuous mode.
• New: Force "avoid edges" to "true" for line labels, since in tile mode they'll always clip poorly at tile boundaries.
• Remove unused "withinPlus0/inside" logic.

In terms of performance, for continuous-rendered maps that don't care, this is an extra boolean-ish check for each symbol at placement time -- should be pretty negligible. For tiled maps, there's an extra check against the tile boundaries but it doesn't involve much extra projection and it might even be a performance win in that it will allow more collision-detection to early-exit near tile boundaries.

Open questions:

• Is it OK to force avoidEdges: true for line labels? I'm counting on MapMode::Tile being a special case (i.e. api-gl) where this makes sense.
• What further testing do we need? The tile-mode render test is the only real testing I've done of this because I don't have an easy way to run an api-gl staging version.
• Am I right that the "withinPlus0" logic still doesn't do anything even with this re-implemented?
• Am I leaving a trap for future maintainers by implying that it's easy to check against tile boundaries without building in strong checking that limits the logic to tiled (i.e. 0-pitch/0-bearing) mode? Maybe just add an assert in isInsideTile?

cc @ansis @ian29 @lilykaiser

[ChrisLoer]

Before

After

[ChrisLoer]

Tests are failing because we need ignores or native port for mapbox/mapbox-gl-js#6961

[ansis]

Is it OK to force avoidEdges: true for line labels? I'm counting on MapMode::Tile being a special case (i.e. api-gl) where this makes sense.

Yes. I'm pretty sure we used to do this

What further testing do we need? The tile-mode render test is the only real testing I've done of this because I don't have an easy way to run an api-gl staging version.

I'm not sure. Would it be useful to have a simple tile server we can run without all of api-gl?

Am I right that the "withinPlus0" logic still doesn't do anything even with this re-implemented?

It looks like it changes the behaviour for tiled rendering when a feature has avoidEdges and is right on the right boundary of the tile (when x === EXTENT). It looks like this was introduced for #6670. But I'm not sure that was the right fix.

Your changes seem good to me.

Am I leaving a trap for future maintainers by implying that it's easy to check against tile boundaries without building in strong checking that limits the logic to tiled (i.e. 0-pitch/0-bearing) mode? Maybe just add an assert in isInsideTile?

An assert seems good to me but this doesn't worry me too much.