Attempt to optimize GivensFGFT(Parallel) for L residual being a MatSparse.

It didn't show any enhancement in performances, so the optimization is disabled but kept in code (the compilation constant OPT_UPDATE_SPARSE_L must be defined for the optimization to be enabled).

The optimization is about the product L = S'*L*S calculation. And the optimization attempt is similar to what have been done for the case where L is dense (in this case it works well).

Here are some results:
J=256, t=128
with "optimization"
>>python$ for i in {1..10}; do python3 test_eigtj_mira_lap_sparse.py;done | grep time | awk '{print $2}'
0.4260462010000001
0.4810052160000007
0.4244757539999995
0.4957183700000005
0.4338328650000003
0.4226386240000002
0.4573114149999995
0.4332176910000003
0.4304637250000001
0.4894571430000001

without "optimization"
>>python$ for i in {1..10}; do python3 test_eigtj_mira_lap_sparse.py;done | grep time | awk '{print $2}'
0.4188697339999994
0.415433846
0.41125124199999963
0.41828689799999985
0.4699177429999999
0.4142650020000005
0.46411832699999955
0.42669971800000006
0.4497221189999996
0.4138595669999994

other runs with J=32, t=2
with "optimization"
>>python$ for i in {1..10}; do python3 test_eigtj_mira_lap_sparse.py;done | grep time | awk '{print $2}'
2.7853622209999997
2.8483473470000007
3.12851224
2.8791390860000003
2.84674213
2.8365956349999992
2.7762758190000003
2.85138975
2.8108322590000006
2.9169309619999995

without "optimization"
>>python$ for i in {1..10}; do python3 test_eigtj_mira_lap_sparse.py;done | grep time | awk '{print $2}'
2.787849833
2.8399939510000003
2.809797146
2.8418070140000005
2.7759090029999998
2.7810152440000007
2.833329901
2.772917798
2.846831988000001
2.765091783