FIx bug density_matrix and torch

[rmoyard]

Context:
When using torch returning the density matrix yields wrong format results.

Description of the Change:
The Torch _asarray function in default.qubit.torch is always unrolling the tensor. I created a specific case for density matrix.

Benefits:
qml.density_matrix works with Torch interface.

Related GitHub Issues:
Closes #2553

[github-actions]

Hello. You may have forgotten to update the changelog!
Please edit doc/releases/changelog-dev.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.

[rmoyard]

[sc-19818]

[codecov]

Codecov Report

Merging #2643 (297ce60) into master (89439a8) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #2643   +/-   ##
=======================================
  Coverage   99.58%   99.58%           
=======================================
  Files         248      248           
  Lines       19972    19975    +3     
=======================================
+ Hits        19889    19892    +3     
  Misses         83       83           

Impacted Files	Coverage Δ
pennylane/_qubit_device.py	98.78% <ø> (ø)
pennylane/devices/default_qubit_torch.py	92.30% <100.00%> (+0.22%)	⬆️

---

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 89439a8...297ce60. Read the comment docs.

[eddddddy]

Hi @rmoyard, looks good to me! Just a minor change and a question, then should be good to go.

[antalszava]

@rmoyard this is awesome! 🥇 🎉

There was a major thing I wondered about related to the change in logic, though the fix seems to overall check out. 👍

[antalszava]

Looks good 👍