perf: HVP with in-place gradient + inner preparation

[gdalle]

Fixes #86 by allowing preparation of the inner gradient in a forward-over-reverse HVP

• Define new trait inner_preparation_behavior with three modalities:
  • PrepareInnerSimple (for finite difference backends)
  • PrepareInnerOverload (for ForwardDiff)
  • DontPrepareInner (for everyone else at the moment, but can evolve)
• Restructure context type hierarchy and add a PrepContext for passing a preparation object created on the correct input type (it cannot be passed as a Cache because some of its values may matter):
  • ForwardDiff will keep it as-is
  • Enzyme will Duplicate it
• Split hvp logic depending on
  • whether the outer backend supports in-place functions (determines whether to use shuffled_gradient or shuffled_gradient!)
  • what type of inner_preparation_behavior the outer backend has
• Define overloaded_input for ForwardDiff and PolyesterForwardDiff to prepare the inner gradient on the right input type
• Fix Cache handling for PolyesterForwardDiff (was incorrectly advertised as supported until now)
• Add more tests, including on a new AutoForwardFromPrimitive

[codecov]

Codecov Report

Attention: Patch coverage is 99.02913% with 2 lines in your changes missing coverage. Please review.

Project coverage is 97.19%. Comparing base (6ae9532) to head (923eca4).

Files with missing lines	Patch %	Lines
DifferentiationInterface/src/second_order/hvp.jl	97.26%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #743      +/-   ##
==========================================
- Coverage   97.45%   97.19%   -0.26%     
==========================================
  Files         125      128       +3     
  Lines        6556     6678     +122     
==========================================
+ Hits         6389     6491     +102     
- Misses        167      187      +20     

Flag	Coverage Δ
DI	97.84% <99.02%> (-0.39%)	⬇️
DIT	95.74% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

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

🚀 New features to boost your workflow:

• ❄ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.