Due to differenced redraw, midpoint.ways was getting stale,
so sometimes dragging a midpoint didn't add the new node to
all ways. Better to calculate participating ways when needed.
This also simplifies the code for adding a midpoint via double
click. It does make filtering midpoints on redraw a bit more
expensive, but a cursory profile didn't show it being a problem.
Fixes#763.
When recalculating parent ways/relations during rebase, a
graph should not add modified or deleted entities as parents.
Such entities will already be correctly marked as parents or
not.
Graph had the correct behavior for deleted entities, but not for
modified entities. This had the effect that if you split a way
that was partially off screen, and then panned so that the way was
re-retrieved, Graph#rebase would mistakenly add back the original
way as a parent of all the nodes that were split into the new
section, making them appear as shared.
Fixes#751.
Multipolygon relations report their geometry as 'area' and are rendered
as such. However, they do not render a stroke. The stroke rendering
will come from the individual lines, which are given the tag
classes of their parent relations, allowing them to have a stroke
style matching the style of simple areas with the same tags.
Untagged circular ways are no longer considered areas. This prevents
an untagged inner way of a multipolygon from rendering as an area and
is consistent with how P2 and JOSM treat them.
In the CSS, it's no longer necessary to deal with multipolygons
explicitly in selectors. But keep in mind that area boundaries can
now be rendered either as lines or as area strokes. In most cases
the selector should be `path.stroke.tag-_____`, i.e. an explicit
`.area` or `.line` classes should not be included.
Finally, the parent ways of selected multipolygons are given the 'selected'
class.
