Rework maxNativeZoom

As @jkuebart pointed out:

It's interesting to note that by using maxNativeZoom as implemented by GridLayer, vector tiles will not be re-rendered at higher zoom levels. Consequently, the stroke width is scaled up along with the tile, while with the previous solution given above the stroke width remained constant.

If maxNativeZoom is only used because there is sufficient detail at a particular zoom level, this might not be the intended effect. If the stroke width should remain visually constant across all zoom levels, a vector tile layer might still need to reimplement getTileSize() and createTile() accordingly.

I think this is a valid concern, and deserves its own issue.

A tentative approach would be to:

• Override L.GridLayer._clampZoom and replace it with an identity function
• Calculate the "data tile" to be fetched for a given "display tile"
• Multiply pxPerExtent by (2 to the power of the zoom delta)
• Calculate some pxOffset for when the "data tile" and "display tile" are not aligned at their top left corners
• Use that pxOffset in all coordinate calculations in PointLayer, PolylineLayer and PolygonLayer

Another approach would be:

• Keep the current _clampZoom logic
• On a zoomend event, iterate through all the tiled renderers (there's a reference in this._vectorTiles) and
  • Reset the size of the containers of the tiled renderers by a factor of (2 to the power of zoom delta)
  • Trigger an update of the tiled renderers

First approach doesn't sound too easy, and might need re-requesting tiles. Second approach sounds simpler, almost too simple.

Start by reading the current GridLayer._clampZoom behavior and the PointLayer, PolylineLayer, and PolygonLayer coordinate calculations. Compare the two proposed approaches around zoomend, this._vectorTiles, data tiles, and display tiles. Done means maxNativeZoom supports higher display zooms while preserving visually constant stroke widths and correct tile alignment.