Update VJP and adjoint-jacobian methods

[maliasadi]

Context:
This PR includes the updates for VJP and AdjDiff in a way that the AdjointJacobian method is more efficient (with reducing the cost of finding trainable params in the list of operations and updating jacobian matrix cache-efficiently) and VJP methods return processing functions required to compute the VJPs.

Description of the Change:

• Add JacobianData to encapsulate the data required in AdjDiff and VJP methods.
• Update the implm. of adjointJacobian and add adjointJacobianJD with a new optimization.
• Update the C++ implm. of vectorJacobianProduct using lambda functions.
• Update the python bindings for adjointJacobian and VJP methods in lightning.qubit.
• Update C++ and Python tests.

Benchmark:
The new adjoint-jacobian is up to 6% faster than the current implementation for large enough circuits.

[github-actions]

Hello. You may have forgotten to update the changelog!
Please edit .github/CHANGELOG.md with:

• A one-to-two sentence description of the change. You may include a small working example for new features.
• A link back to this PR.
• Your name (or GitHub username) in the contributors section.

[codecov]

Codecov Report

Merging #222 (680ee5c) into master (a01bf62) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master     #222   +/-   ##
=======================================
  Coverage   99.71%   99.71%           
=======================================
  Files           4        4           
  Lines         351      354    +3     
=======================================
+ Hits          350      353    +3     
  Misses          1        1           

Impacted Files	Coverage Δ
pennylane_lightning/_version.py	100.00% <100.00%> (ø)
pennylane_lightning/lightning_qubit.py	99.62% <100.00%> (+<0.01%)	⬆️

---

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 a01bf62...680ee5c. Read the comment docs.

[github-actions]

Test Report (C++) on Ubuntu

       1 files  ±  0         1 suites  ±0   0s ⏱️ ±0s
   296 tests ±  0     296 ✔️ ±  0  0 💤 ±0  0 ❌ ±0 
2 687 runs   - 62  2 687 ✔️  - 62  0 💤 ±0  0 ❌ ±0 

Results for commit 680ee5c. ± Comparison against base commit a01bf62.

♻️ This comment has been updated with latest results.

[chaeyeunpark]

Hi Ali, nice work 👍 I here just drop several comments.

[mlxd]

Great work on this @maliasadi
I have some questions and thoughts. I reckon we can figure out the integration testing mentioned earlier and identify how to best examine the use with PL.

Maybe @josh146 would have an idea how to best do some runtime and integration testing with/against the Python-only implementation.

[chaeyeunpark]

Hi Ali, I just wanna approve this PR. Great job 👍

[mlxd]

Excellent work on this @maliasadi
Happy to approve too. Once the PL changes come we can do the integration.