Make decompression more efficient

[gdalle]

Source

• Refactor the result types into NonSymmetricColoringResult, StarSetColoringResult and TreeSetColoringResult, plus the ugly LinearSystemColoringResult.
• In every result type, make sure that S isa SparseMatrixCSC (fix Make S::SparseMatrixCSC{Bool,Int} #69) and put the necessary information for allocation-less decompression into dense and sparse A. The complexity is now mainly in the constructors for these result types.
• Add a decompression_eltype keyword argument to coloring (fix Provide eltype of decompression target A #44). Use it inside the TreeSetColoringResult constructor to initialize stored_values.
• Move the decompression type parameter from the ColoringProblem to the GreedyColoringAlgorithm (fix Should the decompression method go in the ColoringProblem or in the ColoringAlgorithm? #66).
• Warn that :symmetric structures require a :column partition (fix Add more methods for coloring? #51).

Tests

• Add type stability and performance checks for all colorings and all decompressions (only into SparseMatrixCSC for now).
• Test LinearSystemColoringResult in addition to the native result

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.37%. Comparing base (cb1c2c6) to head (24d93a4).

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #71      +/-   ##
==========================================
- Coverage   99.83%   98.37%   -1.47%     
==========================================
  Files          11       10       -1     
  Lines         605      614       +9     
==========================================
  Hits          604      604              
- Misses          1       10       +9     

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