* `AX_LIB_EXPAT` has a big practical issue: It tries to detect expat
  by injecting various paths, among which is /usr/lib. This is extremely
  problematic, as /usr/lib usually contains 32-bit libraries on 64-bit
  systems and thereby kills linkers such as LLD that reject 32-bit
  binaries when linking 64-bit binaries. `pkg-config` does not inject
  /usr/lib, and can also be overridden at configure time.

Closes: https://bugs.gentoo.org/737974

Increase Boost Python function arity only when necessary

See merge request count0/graph-tool!36

Fix dot graph reader

Closes #450

See merge request count0/graph-tool!35

Adjust use of libgvc to fix two graphviz_draw issues

Closes #597

See merge request count0/graph-tool!34

The correct type for the dynamic property map key is actually
graph_property_tag, not multigraph_t*. It looks like graph property maps
only store a single value...

One issue was reported to the mailing list on 2020-08-17 by Rolf
Sander; the testcase is:

import graph_tool.all as gt
g = gt.Graph()
v0 = g.add_vertex()
v1 = g.add_vertex()
e01 = g.add_edge(v0,v1)
e11 = g.add_edge(v1,v1) # loop
gt.graphviz_draw(g)

The other is issue #597 "Graphviz_draw core dumping when using HTML labels"

Basically these changes add extra type information for interacting
with libgvc.

Remove uncessary exception throwing and property map checking.

UTF8 is not strictly valid for GML, but it doesn't hurt.

With this, edge properties with convertible types are always copied,
regardless if the edge set is the same between target and source
graphs. In case the edge set is not the same, the properties are copied
only for the edges in the intersection.

This fixes issue #668

Use lambdas and pack expansion to reduce template instantiations in action_dispatch

This change produces significant reductions in compiler memory use (4.15GB to 2.81GB in my `graph_assortativity` testcase) and compiler runtime (180m to 168m overall, 206s to 188s in the testcase) by:

1. Performing type iteration with parameter pack expansion where possible
2. Testing the stored `boost::any` types sequentially instead of all at once (M+N instead of M*N operations)

I see no significant change in runtime with this approach, though it should in theory be an improvement.

See merge request !33

Improve build times through header removal and relocation

See merge request !31

- Eliminate unused headers
- reduce dependencies among headers (slightly)

The include-what-you-use tool (IWYU) was used to analyze header
dependencies and recommend fixes; some additional manual work was
then performed.

Add missing include

Ensure supplied position map for planar_layout is owned by view

See merge request !32