Customize quantum_info functions

[BrunoLiegiBastonLiegi]

This provides customization for the quantum_info functions introduced in qiboteam/qibo#1583.
In the following some benchmarks run on 10 repetions:

pauli basis

### 3 qubits
--> numpy:  0.0009936720016412437
--> numba:  0.05624490500122192
--> cupy:  0.13221715999679873
### 5 qubits
--> numpy:  0.24638988500009873
--> numba:  0.06344797999918228
--> cupy:  0.005442648998723598
### 7 qubits
--> numpy:  82.37901496199993
--> numba:  15.51170543400076
--> cupy:  0.7553486539982259

random gaussian

##### random gaussian
### 5 qubits
--> numpy:  0.0017942120011866791
--> numba:  0.0047024770010466455
--> cupy:  0.13087414899928262
### 10 qubits
--> numpy:  0.35412388399890915
--> numba:  0.036867562999759684
--> cupy:  0.0028848960009781877
### 12 qubits
--> numpy:  6.0999391290006315
--> numba:  0.8496762070008117
--> cupy:  0.002546637000705232

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 100.00%. Comparing base (193b569) to head (8113c44).

Additional details and impacted files

@@            Coverage Diff            @@
##              main      #212   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files            7         7           
  Lines          604       615   +11     
=========================================
+ Hits           604       615   +11     

Flag	Coverage Δ
unittests	100.00% <100.00%> (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.

[BrunoLiegiBastonLiegi]

This is ready for a first round of reviews. The notably missing elements are:

• jitting the system ordered functions, this is not possible right now as numba does not support the construction of tuples with length not known at compile time, which is needed in the current implementation of vectorization/unvectorization_system. A workaround, may be possible but I did not spend too much time looking into that.
• jitting random_unitary and the functions that depend on it was not possible due to the scipy.linalg.expm method not working with numba. We could implement the matrix exponential ourselves directly in native numpy, for istance through the padet method, but again did not spend too much time investigating that.
• there are a bunch of minor warnings raised by numba concerning the use of parallel=True in functions where no parallelization was possible, and multiplication of fortran and c ordered arrays being suboptimal, which I could look into to get rid of.