Allow to choose the quadrature degree used to evaluate the
expressions in the plugin integrate. To do so, the token
quad-deg needs to be added to the input file in the integrate
plugin section. The default quadrature type is taken from that used
to set the solution points. If the user adds
quad-pts-{etype} = quad_name to the integrate plugin
section, then the the quad_name quadrature will be used.

Optimize the performance by multiplying M15 by M12
reducing the sike of the matrix-matrix kernel

Two optimizations:

1) Avoid multiplications of smats * flux that are not needed.
Internal flux points and unique internal flux points coincide
in tensor product elements

2) Reduce the size of the tdivtpcorf kernel (contribution of
the internal flux points to the divergence)

Such optimizations can also be applied (in a more complex
way) to pri elements.

Improve the performance of the low-storage pseudo stepper.

As shown in:

TODO: some improvements can be done with
jacobi_norm vs jacobi in the internal basis.

As shown in

Efficient Equilibrated Flux Reconstruction in High Order Raviart-
Thomas Space for Discontinuous Galerkin methods

This drastically reduces the condition number of the VDM
matrix

Remove max matrix size limitation for libxsmm.

TODO: change name

Use fr-sd = True option in [solver]

Tet implementation is not finished, divergence needs to be implemented.

1) only_root_cell allows to speed up the computation of the RHS
Jacobian for Fourier analysis, as only the RHS Jacobian of the elements
belonging to the root pattern are computed

2) Output  to binary to speed up the reading of these matrices

Python 3.10 support.

Extend the CUDA error mapping table to handle 'no device' errors.

Cleanup stream/queue names in the backend.

Cleanups.

This allows to perform an a posteriori Fourier analysis
using the Jacobian matrix and the position of the
solution points