Optimize the input state-vector copy into the LGPU

[LuisAlfredoNu]

Context:
After running different algorithm with LGPU and perform a memory profile. Show a memory bottleneck for LGPU on the Python layer because the peak of memory is 3 times the need for the computation.

Description of the Change:
Remove tmp allocation and skip indexes computation for common cases.

• Remove temporal GPU allocation for input values and indexes.
• The input state vector is copied directly from the host if the target wires are contiguous and start in the most/least significant wires (which are the most common cases).
• In the case of custom target wires, LGPU follow the previous algorithm but with a speedup in the index computation thought parallel computing with OpenMP

Benefits:
Using a test algorithm with 31 qubits produce the following memory profile:

Reduction of the memory peak from 100GB to 66GB

Note: memray measures all the memory allocation, even for the GPU cudaMallocX.

Using the following toy circuit

   state_init =  random_normalize_sv(wires-1)
   target_wires = wires[:-1]
   dev = qml.device("lightning.gpu", wires=wires)

    def circuit():
        qml.StatePrep(input_state, wires=target_wires)
                
        return qml.expval(qml.PauliZ(0))

Produce the following times

Possible Drawbacks:

Related GitHub Issues:
[sc-58833]

[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

Attention: Patch coverage is 96.59091% with 3 lines in your changes missing coverage. Please review.

Project coverage is 98.11%. Comparing base (109db9f) to head (901be60).
Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
...lightning/core/src/utils/cuda_utils/DataBuffer.hpp	83.33%	2 Missing ⚠️
pennylane_lightning/lightning_gpu/_state_vector.py	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1071      +/-   ##
==========================================
+ Coverage   97.99%   98.11%   +0.12%     
==========================================
  Files         233      232       -1     
  Lines       40019    39268     -751     
==========================================
- Hits        39215    38527     -688     
+ Misses        804      741      -63     

☔ 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.

[AmintorDusko]

Good job! A great improvement for memory usage. I assume you checked the codecov warning and they are all fake, right?

[multiphaseCFD]

Nice one. Thanks @LuisAlfredoNu

[AmintorDusko]

Thank you for your nice job!

[maliasadi]

Awesome work! Thanks @LuisAlfredoNu 🥇

Don't forget to add your PR to the changelog :)

[maliasadi]

Happy to approve 🥳