Switch to DifferentiationInterface

[gdalle]

• Bump version to 0.6.0

Breaking changes

• Bump Julia compat to 1.10 lower bound
• Remove support for higher order arrays as x and y: only vectors since we would need to reshape anyway
• Replace AbstractDifferentiation with DifferentiationInterface for condition backend specification
• Remove the AbstractLinearSolver construction in favor of a callable with two methods (A, b) -> s and (A, B) -> S
• Separate the condition backends into x and y (useful if they have different dimensions)

Non-breaking changes

• Implement a dense version of the operators which allocates the full jacobian
• Implement batched solves, only for dense version (see block_gmres fails with LinearOperator JuliaSmoothOptimizers/Krylov.jl#854)
• Base the dense and lazy operators on DifferentiationInterface, both in forward and in split reverse mode
• Add rules for Enzyme in (batch) forward mode
• Add projection step to ChainRules rrule
• Update docs and tests
• Remove outdated and unused benchmarks
• Remove precompilation

[codecov]

Codecov Report

Attention: Patch coverage is 94.92754% with 7 lines in your changes are missing coverage. Please review.

Project coverage is 95.48%. Comparing base (b70edec) to head (4b023f3).

Files	Patch %	Lines
src/operators.jl	93.47%	6 Missing ⚠️
src/implicit_function.jl	94.44%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #135      +/-   ##
==========================================
- Coverage   97.34%   95.48%   -1.86%     
==========================================
  Files           9        6       -3     
  Lines         188      155      -33     
==========================================
- Hits          183      148      -35     
- Misses          5        7       +2     

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