Support user-defined set types

[adrhill]

Closes #23 by supporting all sets of type AbstractSet{<:Integer}.
Breaking since the current pattern API now requires the user to specify the set type in the tracer:

julia> using SparseConnectivityTracer

julia> x = rand(3);

julia> f(x) = [x[1]^2, 2 * x[1] * x[2]^2, sin(x[3])];

julia> pattern(f, JacobianTracer{BitSet}, x) # <--- Set type needs to be specifies
3×3 SparseArrays.SparseMatrixCSC{Bool, UInt64} with 4 stored entries:
 1  ⋅  ⋅
 1  1  ⋅
 ⋅  ⋅  1

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 53.12500% with 45 lines in your changes are missing coverage. Please review.

Project coverage is 70.49%. Comparing base (b1f2ec3) to head (76f9d7c).

Files	Patch %	Lines
src/tracers.jl	57.77%	19 Missing ⚠️
src/conversion.jl	36.84%	12 Missing ⚠️
src/overload_jacobian.jl	27.27%	8 Missing ⚠️
src/overload_hessian.jl	40.00%	6 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #31      +/-   ##
==========================================
- Coverage   76.32%   70.49%   -5.83%     
==========================================
  Files           8        8              
  Lines         245      261      +16     
==========================================
- Hits          187      184       -3     
- Misses         58       77      +19     

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

[gdalle]

You were right it's too complex for me to review so I just flagged a few things that stood out to me as abnormal

What is the point of these reference tests if the reference files are changed automatically?

[adrhill]

To highlight the difference in performance on large scale problems:

julia> x = rand(220, 220, 3, 1); # WHCN

julia> w = rand(5, 5, 3, 8); # Conv((5, 5), 3 => 8)

julia> @time pattern(x -> conv(x, w), JacobianTracer{BitSet}, x)
 57.282859 seconds (141.11 M allocations: 655.483 GiB, 57.00% gc time, 0.14% compilation time)
373248×145200 SparseMatrixCSC{Bool, UInt64} with 27993600 stored entries:
...

julia> @time pattern(x -> conv(x, w), JacobianTracer{Set{UInt64}}, x)
 20.156197 seconds (303.10 M allocations: 45.897 GiB, 14.60% gc time, 0.44% compilation time)
373248×145200 SparseMatrixCSC{Bool, UInt64} with 27993600 stored entries:
...

For small and medium size problems, BitSet is more performant.