adding fixed locations to sfdp

[claytonpbarrows]

I commonly use SFDP to determine the locations of a subset of vertices when some locations are known. This PR implements this functionality by adding the fixed kwarg.
If this is too niche, please let me know and I'll close the PR.

[codecov]

Codecov Report

Merging #29 (8c45251) into master (00491f9) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master      #29   +/-   ##
=======================================
  Coverage   98.62%   98.62%           
=======================================
  Files           7        7           
  Lines         363      364    +1     
=======================================
+ Hits          358      359    +1     
  Misses          5        5           

Impacted Files	Coverage Δ
src/sfdp.jl	98.11% <100.00%> (+0.03%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 00491f9...8c45251. Read the comment docs.

[hexaeder]

I don't think this is too niche, fixed positions for Spring and SFDP layouts are definitely on my feature wishlist for the next release. However I've recently started quite a big refactor of this package in #27 (wip docs), I think we should finish this process first before adding new features. Comments on this new interfaces are allways welcome!

The main idea is to split the parameters of each algorithm from it's actual application to a graph. The main problem here is to keep the algorithm parametrization as general as possible, for example, it should be possible to add more nodes and edges to the graph without need of reparametrizing the algorithm. I'd therefore suggest to implement this as an Dict{Int, Point{dim, Ptype}} instead of a BitVector and initial positions of fixed size.

layout = SFDP(; fixed=Dict(1=>Point(0,0), 3=>Point(1,1))
positions = layout(adj_matrix)

Would still work with graphs of different sizes by filling up all of the other field with random positions. This comes quite handy if you don't wan't to think about the initialization of all of the other start positions.
I becomes quite unelegant in the presence of both (initial positions and fixed positions) though, where the fixed positions should take precedence....

[claytonpbarrows]

OK, once you have merged #27, I'll rebase and address your comment.

[hexaeder]

just wanted to bump this because 0.4 is released now