fix(FiniteDiff): use fdhtype for hvp to match hessian accuracy

[bdrhill]

Summary

• Implements custom hvp, hvp!, gradient_and_hvp, and gradient_and_hvp! for AutoFiniteDiff
• Uses fdhtype (default Val(:hcentral)) for the Hessian computation instead of the generic JVP-of-gradient fallback which used fdtype (default Val(:forward))
• The forward-over-forward finite difference composition caused significant precision loss

Fixes #1012

Test plan

• Verify with the MWE from Bug(FiniteDiff): hvp accuracy under AutoFiniteDiff() defaults disagrees with hessian #1012:

  using DifferentiationInterface, FiniteDiff, ADTypes
  
  f(x) = sum(x .^ 2)
  x = [1.0, 2.0, 3.0]
  v = [1.0, 0.0, 0.0]
  backend = AutoFiniteDiff()
  
  H = hessian(f, backend, x)
  hv = hvp(f, backend, x, (v,))[1]
  @assert isapprox(hv, H * v, rtol=1e-6)  # should pass now

• Run FiniteDiff backend tests

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.15%. Comparing base (cbfeb7c) to head (fede0bd).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1022      +/-   ##
==========================================
+ Coverage   98.14%   98.15%   +0.01%     
==========================================
  Files         138      138              
  Lines        8131     8194      +63     
==========================================
+ Hits         7980     8043      +63     
  Misses        151      151              

Flag	Coverage Δ
DI	98.87% <100.00%> (+0.01%)	⬆️
DIT	96.22% <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.