Fixes vector-Hessian product bug

[josh146]

Context: The existing parameter-shift Hessian implementation returns incorrect Hessians for cost functions that perform post-processing on a vector-valued QNode.

Description of the Change:

• The bug was fixed in all three main interfaces (autograd, PyTorch, TF), by making sure that the vector-Hessian product is divided by the norm squared of dy.

• Tests were added for all three interfaces to ensure that this bug is fixed.

Benefits: Fixes the aforementioned bug.

Possible Drawbacks: n/a

Related GitHub Issues: Fixes #1424

[codecov]

Codecov Report

Merging #1436 (572443c) into master (fe71ae0) will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #1436   +/-   ##
=======================================
  Coverage   98.24%   98.24%           
=======================================
  Files         160      160           
  Lines       12016    12016           
=======================================
  Hits        11805    11805           
  Misses        211      211           

Impacted Files	Coverage Δ
pennylane/interfaces/tf.py	97.43% <ø> (ø)
pennylane/interfaces/torch.py	100.00% <ø> (ø)
pennylane/interfaces/autograd.py	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update fe71ae0...572443c. Read the comment docs.

[albi3ro]

Thanks!

Reproduced it locally and the bug is gone.

Can't see anything particular to comment on.