Store graph in result to allow generic matrices

[gdalle]

Fix #126, fix #131

Before this PR, the result type would always store a SparseMatrixCSC and forget the initial matrix A. Now it stores both the initial matrix A and the AdjacencyGraph or BipartiteGraph used for coloring, which contains the sparse conversion. This has two main advantages:

• To prepare for row decompression we do not need to recompute a transpose, it is already done in the BipartiteGraph
• When decompressing out-of-place, we can instantiate similar(A) instead of similar(S), which makes more sense and paves the way for structured matrices

I also remove some eltype constraints on A and B which were not necessary. For instance, if eltype(B) == Float32 and eltype(A) == Float64, we can decompress from B to A without issue.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 100.00%. Comparing base (a03a455) to head (8c073ee).
Report is 4 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff            @@
##              main      #133   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           11        11           
  Lines          851       870   +19     
=========================================
+ Hits           851       870   +19     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[github-actions]

Benchmark Results

	main	12dcf60...	main/12dcf60e027c12...
coloring/nonsymmetric/column/direct/n=1000/p=0.002	0.0861 ± 0.004 ms	0.0858 ± 0.0041 ms	1
coloring/nonsymmetric/column/direct/n=1000/p=0.005	0.187 ± 0.0075 ms	0.189 ± 0.008 ms	0.991
coloring/nonsymmetric/column/direct/n=1000/p=0.01	0.391 ± 0.013 ms	0.396 ± 0.013 ms	0.987
coloring/nonsymmetric/column/direct/n=100000/p=0.0001	0.068 ± 0.0053 s	0.0751 ± 0.0023 s	0.905
coloring/nonsymmetric/column/direct/n=100000/p=2.0e-5	12.4 ± 0.3 ms	13.5 ± 0.47 ms	0.916
coloring/nonsymmetric/column/direct/n=100000/p=5.0e-5	29.7 ± 1.1 ms	0.0328 ± 0.00089 s	0.905
coloring/nonsymmetric/row/direct/n=1000/p=0.002	0.102 ± 0.0047 ms	0.0883 ± 0.0042 ms	1.15
coloring/nonsymmetric/row/direct/n=1000/p=0.005	0.213 ± 0.0085 ms	0.192 ± 0.0078 ms	1.11
coloring/nonsymmetric/row/direct/n=1000/p=0.01	0.44 ± 0.015 ms	0.401 ± 0.014 ms	1.1
coloring/nonsymmetric/row/direct/n=100000/p=0.0001	0.075 ± 0.004 s	0.076 ± 0.002 s	0.986
coloring/nonsymmetric/row/direct/n=100000/p=2.0e-5	14.9 ± 0.39 ms	13.7 ± 0.45 ms	1.09
coloring/nonsymmetric/row/direct/n=100000/p=5.0e-5	0.0346 ± 0.00096 s	0.033 ± 0.00055 s	1.05
coloring/symmetric/column/direct/n=1000/p=0.002	0.271 ± 0.015 ms	0.272 ± 0.015 ms	0.996
coloring/symmetric/column/direct/n=1000/p=0.005	0.734 ± 0.028 ms	0.733 ± 0.028 ms	1
coloring/symmetric/column/direct/n=1000/p=0.01	1.77 ± 0.053 ms	1.79 ± 0.056 ms	0.988
coloring/symmetric/column/direct/n=100000/p=0.0001	0.471 ± 0.019 s	0.397 ± 0.012 s	1.19
coloring/symmetric/column/direct/n=100000/p=2.0e-5	0.0365 ± 0.0023 s	0.0369 ± 0.0008 s	0.991
coloring/symmetric/column/direct/n=100000/p=5.0e-5	0.127 ± 0.0069 s	0.123 ± 0.0042 s	1.03
coloring/symmetric/column/substitution/n=1000/p=0.002	0.609 ± 0.031 ms	0.613 ± 0.03 ms	0.994
coloring/symmetric/column/substitution/n=1000/p=0.005	1.61 ± 0.062 ms	1.61 ± 0.059 ms	1
coloring/symmetric/column/substitution/n=1000/p=0.01	3.41 ± 0.096 ms	3.46 ± 0.1 ms	0.985
coloring/symmetric/column/substitution/n=100000/p=0.0001	0.758 ± 0.017 s	0.817 ± 0.047 s	0.927
coloring/symmetric/column/substitution/n=100000/p=2.0e-5	0.0957 ± 0.01 s	0.102 ± 0.017 s	0.94
coloring/symmetric/column/substitution/n=100000/p=5.0e-5	0.292 ± 0.019 s	0.338 ± 0.037 s	0.865
decompress/nonsymmetric/column/direct/n=1000/p=0.002	3.94 ± 0.25 μs	4.68 ± 0.31 μs	0.842
decompress/nonsymmetric/column/direct/n=1000/p=0.005	8.35 ± 0.67 μs	9.88 ± 0.87 μs	0.845
decompress/nonsymmetric/column/direct/n=1000/p=0.01	26.7 ± 5.9 μs	21.1 ± 6.9 μs	1.26
decompress/nonsymmetric/column/direct/n=100000/p=0.0001	5.4 ± 0.36 ms	5.79 ± 0.51 ms	0.933
decompress/nonsymmetric/column/direct/n=100000/p=2.0e-5	1.09 ± 0.19 ms	1.15 ± 0.15 ms	0.953
decompress/nonsymmetric/column/direct/n=100000/p=5.0e-5	2.72 ± 0.27 ms	2.79 ± 0.29 ms	0.973
decompress/nonsymmetric/row/direct/n=1000/p=0.002	4.08 ± 0.28 μs	4.86 ± 0.4 μs	0.84
decompress/nonsymmetric/row/direct/n=1000/p=0.005	7.68 ± 0.45 μs	9.21 ± 0.55 μs	0.834
decompress/nonsymmetric/row/direct/n=1000/p=0.01	14.8 ± 0.95 μs	17.5 ± 1.5 μs	0.849
decompress/nonsymmetric/row/direct/n=100000/p=0.0001	2.15 ± 1.4 ms	2.09 ± 0.17 ms	1.03
decompress/nonsymmetric/row/direct/n=100000/p=2.0e-5	0.385 ± 0.044 ms	0.434 ± 0.074 ms	0.887
decompress/nonsymmetric/row/direct/n=100000/p=5.0e-5	0.922 ± 0.11 ms	1.03 ± 0.18 ms	0.897
decompress/symmetric/column/direct/n=1000/p=0.002	3.74 ± 0.24 μs	4.75 ± 0.42 μs	0.788
decompress/symmetric/column/direct/n=1000/p=0.005	7.35 ± 0.42 μs	9.19 ± 0.62 μs	0.8
decompress/symmetric/column/direct/n=1000/p=0.01	15.3 ± 0.96 μs	18.5 ± 1.5 μs	0.83
decompress/symmetric/column/direct/n=100000/p=0.0001	4.37 ± 0.17 ms	4.67 ± 0.37 ms	0.937
decompress/symmetric/column/direct/n=100000/p=2.0e-5	0.847 ± 0.1 ms	0.926 ± 0.13 ms	0.915
decompress/symmetric/column/direct/n=100000/p=5.0e-5	2.1 ± 0.22 ms	2.33 ± 0.67 ms	0.905
decompress/symmetric/column/substitution/n=1000/p=0.002	0.0644 ± 0.0043 ms	0.065 ± 0.0041 ms	0.991
decompress/symmetric/column/substitution/n=1000/p=0.005	0.159 ± 0.0077 ms	0.161 ± 0.0074 ms	0.983
decompress/symmetric/column/substitution/n=1000/p=0.01	0.338 ± 0.014 ms	0.346 ± 0.014 ms	0.976
decompress/symmetric/column/substitution/n=100000/p=0.0001	0.0597 ± 0.0029 s	0.0675 ± 0.0044 s	0.885
decompress/symmetric/column/substitution/n=100000/p=2.0e-5	12.3 ± 0.38 ms	12.7 ± 0.39 ms	0.966
decompress/symmetric/column/substitution/n=100000/p=5.0e-5	28.4 ± 0.74 ms	30.6 ± 0.74 ms	0.928
time_to_load	0.218 ± 0.0025 s	0.218 ± 0.0008 s	1

[amontoison]

Except two small comments, it's LGTM.

[gdalle]

I think both those comments are resolved?

[amontoison]

I think both those comments are resolved?

Tu peux merge si tu veux après