Set Tile loaded/rendered instead of marking tile as optional.

[asheemmamoowala]

Fixes #11957.

When a tile has no features CustomGeometryTile::setTileData marks it as TileNecessity::optional, before calling GeometryTile:setData. This call to setData triggers parse and layout on the GeometryTileWorker and the worker will subsequently update the tile, which requests a new render frame. In the next frame, the tile is marked as TileNecessity::Required again, triggerring a reload of the tile data and subsequent repeat layout of the empty feature data - creating a never-ending loop.

The fix here is to mark the tile as loaded and renderable instead of optional. This way the tile skips layout (and placement) and does not incur future reloads until evicted fro the tile cache.

Verified the fix on iOS, macOS, and Android. CPU consumption drops back to 0 and tiles behave correctly.

The one thing I am not sure of in this approach is using the loaded and renderable protected members from the Tile parent class.

cc @ChrisLoer @fabian-guerra

[ChrisLoer]

It's OK for classes derived from Tile to modify loaded and renderable, but I think the easier way to manage the state here is to call setData with an empty data set -- we already support that pathway, and it will generate a renderable tile with no features in it, using the same code as the rest of GeometryTile. And I don't think you need to change the tile necessity here.

I haven't looked into why all those tests timed out, but it does seem suspicious...

[fabian-guerra]

I tried this fix and is working. Also ran the tests and is timing out. So this change affected somehow the Source.CustomGeometrySourceSetTileData test.

[asheemmamoowala]

The Source.CustomGeometrySourceSetTileData was legitimately failing. The fix suggested by @ChrisLoer fixes it though, so I've pushed that change.

[ChrisLoer]

🚀

[friedbunny]

iOS changelog added in #12086.