Extract diff engine bindings into standalone SparseDiffPy package

[Transurgeon]

Move C bindings (_bindings/) to the new SparseDiffPy package at SparseDifferentiation/SparseDiffPy and import via
from sparsediffpy import _sparsediffengine as _diffengine. Remove the diff_engine_core submodule and its build configuration.

[github-actions]

Benchmarks that have stayed the same:

   before           after         ratio
 [7919d4d5]       [8d00889c]
      791±0ms          807±0ms     1.02  simple_QP_benchmarks.LeastSquares.time_compile_problem
      267±0ms          272±0ms     1.02  matrix_stuffing.ParamSmallMatrixStuffing.time_compile_problem
     14.3±0ms         14.5±0ms     1.02  simple_QP_benchmarks.ParametrizedQPBenchmark.time_compile_problem
      1.33±0s          1.35±0s     1.02  matrix_stuffing.ParamConeMatrixStuffing.time_compile_problem
      1.83±0s          1.86±0s     1.02  simple_QP_benchmarks.UnconstrainedQP.time_compile_problem
      859±0ms          870±0ms     1.01  simple_LP_benchmarks.SimpleScalarParametrizedLPBenchmark.time_compile_problem
      12.5±0s          12.6±0s     1.01  finance.CVaRBenchmark.time_compile_problem
      2.71±0s          2.74±0s     1.01  quantum_hilbert_matrix.QuantumHilbertMatrix.time_compile_problem
      660±0ms          665±0ms     1.01  matrix_stuffing.ConeMatrixStuffingBench.time_compile_problem
      233±0ms          235±0ms     1.01  simple_QP_benchmarks.SimpleQPBenchmark.time_compile_problem
      4.85±0s          4.88±0s     1.01  optimal_advertising.OptimalAdvertising.time_compile_problem
      269±0ms          270±0ms     1.00  slow_pruning_1668_benchmark.SlowPruningBenchmark.time_compile_problem
      21.6±0s          21.6±0s     1.00  sdp_segfault_1132_benchmark.SDPSegfault1132Benchmark.time_compile_problem
     13.8±0ms         13.8±0ms     1.00  simple_LP_benchmarks.SimpleFullyParametrizedLPBenchmark.time_compile_problem
      304±0ms          304±0ms     1.00  gini_portfolio.Yitzhaki.time_compile_problem
      4.73±0s          4.73±0s     1.00  svm_l1_regularization.SVMWithL1Regularization.time_compile_problem
      4.14±0s          4.13±0s     1.00  huber_regression.HuberRegression.time_compile_problem
      932±0ms          928±0ms     1.00  gini_portfolio.Cajas.time_compile_problem
      490±0ms          488±0ms     1.00  semidefinite_programming.SemidefiniteProgramming.time_compile_problem
      10.9±0s          10.8±0s     0.99  simple_LP_benchmarks.SimpleLPBenchmark.time_compile_problem
      1.00±0s          987±0ms     0.99  finance.FactorCovarianceModel.time_compile_problem
      227±0ms          222±0ms     0.98  gini_portfolio.Murray.time_compile_problem
     38.4±0ms         37.6±0ms     0.98  matrix_stuffing.SmallMatrixStuffing.time_compile_problem
      142±0ms          138±0ms     0.98  high_dim_convex_plasticity.ConvexPlasticity.time_compile_problem
      1.57±0s          1.50±0s     0.95  tv_inpainting.TvInpainting.time_compile_problem

[Transurgeon]

@dance858 this PR cleans up the DNLP repo and adds sparsediffpy as a dependency. All tests seem to pass in the CI :).

[dance858]

Awesome! Super nice.