Fix two bugs in c5496442 (pyfaust.poly.poly extension to avoid round trips...

Fix two bugs in c5496442 (pyfaust.poly.poly extension to avoid round trips between wrapper and C++ core when computing the exp action on multiple tau/coeffs in expm_multiply) and manage missing case when X is a vector.

It enhances the performance on GPU: (benchmarks ran on GPU with the new extension -- group_coeffs == True, then on CPU, and finally on GPU with group_coeffs=False).

(test_pyfaust) hhadjdji@v100alpha:~$ python ~/benchmark_expm_multiply.py 10000 5000 1024 150 100 pyfaust_gpu group_coeffs
generating a symmetric sparse matrix...
L.shape (10000, 10000)
L.nnz: 5000
L.density: 5e-05
K= 150
X.shape[1]: 1024
number of tau: 100
group_coeffs= True
expm_multiply poly_meth: 2
FaustCoreCpp polyCoeffsSeq
poly Seq GPU
pyfaust gpu time: 74.51493524294347
expm_multiply poly_meth: 2
FaustCoreCpp polyCoeffsSeq
poly Seq GPU

(test_pyfaust) hhadjdji@x5570node2:~$ python3 benchmark_expm_multiply.py 10000 5000 1024 150 100 pyfaust
generating a symmetric sparse matrix...
L.shape (10000, 10000)
L.nnz: 4997
L.density: 4.997e-05
K= 150
X.shape[1]: 1024
number of tau: 100
group_coeffs= False
expm_multiply poly_meth: 2
pyfaust time: 123.20772746193688
expm_multiply poly_meth: 2

(test_pyfaust) hhadjdji@v100alpha:~$ python ~/benchmark_expm_multiply.py 10000 5000 1024 150 100 pyfaust_gpu
generating a symmetric sparse matrix...
L.shape (10000, 10000)
L.nnz: 4999
L.density: 4.999e-05
K= 150
X.shape[1]: 1024
number of tau: 100
group_coeffs= False
expm_multiply poly_meth: 2
pyfaust gpu time: 380.94207076495513
expm_multiply poly_meth: 2

https://gitlab.inria.fr/faustgrp/faust/-/snippets/714