set compact thread binding for numpy test

[smoors]

thread_binding variable is defined in:

• add BLAS test #268

i also added a launcher parameter for convenience

[casparvl]

Prior numpy test performance numbers (without this change):

[ RUN      ] EESSI_NumPy %scale=1_node %module_name=SciPy-bundle/2023.11-gfbf-2023b /ae66ca75 @snellius:rome+default
[ RUN      ] EESSI_NumPy %scale=1_node %module_name=SciPy-bundle/2023.07-gfbf-2023a /5c365c0d @snellius:rome+default
[ RUN      ] EESSI_NumPy %scale=1_node %module_name=SciPy-bundle/2023.02-gfbf-2022b /4cf00525 @snellius:rome+default
[       OK ] (1/3) EESSI_NumPy %scale=1_node %module_name=SciPy-bundle/2023.02-gfbf-2022b /4cf00525 @snellius:rome+default
P: dot: 3.03 s (r:0, l:None, u:None)
P: svd: 10.14 s (r:0, l:None, u:None)
P: cholesky: 4.31 s (r:0, l:None, u:None)
P: eigendec: 67.03 s (r:0, l:None, u:None)
P: inv: 4.9 s (r:0, l:None, u:None)
[       OK ] (2/3) EESSI_NumPy %scale=1_node %module_name=SciPy-bundle/2023.07-gfbf-2023a /5c365c0d @snellius:rome+default
P: dot: 3.1 s (r:0, l:None, u:None)
P: svd: 29.04 s (r:0, l:None, u:None)
P: cholesky: 4.1 s (r:0, l:None, u:None)
P: eigendec: 148.56 s (r:0, l:None, u:None)
P: inv: 4.52 s (r:0, l:None, u:None)
[       OK ] (3/3) EESSI_NumPy %scale=1_node %module_name=SciPy-bundle/2023.11-gfbf-2023b /ae66ca75 @snellius:rome+default
P: dot: 3.06 s (r:0, l:None, u:None)
P: svd: 44.33 s (r:0, l:None, u:None)
P: cholesky: 6.17 s (r:0, l:None, u:None)
P: eigendec: 143.04 s (r:0, l:None, u:None)
P: inv: 4.62 s (r:0, l:None, u:None)
[----------] all spawned checks have finished

With this change:

[ RUN      ] EESSI_NumPy %scale=1_node %module_name=SciPy-bundle/2023.11-gfbf-2023b /ae66ca75 @snellius:rome+default
[ RUN      ] EESSI_NumPy %scale=1_node %module_name=SciPy-bundle/2023.07-gfbf-2023a /5c365c0d @snellius:rome+default
[ RUN      ] EESSI_NumPy %scale=1_node %module_name=SciPy-bundle/2023.02-gfbf-2022b /4cf00525 @snellius:rome+default
[       OK ] (1/3) EESSI_NumPy %scale=1_node %module_name=SciPy-bundle/2023.02-gfbf-2022b /4cf00525 @snellius:rome+default
P: dot: 3.03 s (r:0, l:None, u:None)
P: svd: 9.4 s (r:0, l:None, u:None)
P: cholesky: 4.15 s (r:0, l:None, u:None)
P: eigendec: 68.34 s (r:0, l:None, u:None)
P: inv: 4.6 s (r:0, l:None, u:None)
[       OK ] (2/3) EESSI_NumPy %scale=1_node %module_name=SciPy-bundle/2023.07-gfbf-2023a /5c365c0d @snellius:rome+default
P: dot: 2.99 s (r:0, l:None, u:None)
P: svd: 31.75 s (r:0, l:None, u:None)
P: cholesky: 3.96 s (r:0, l:None, u:None)
P: eigendec: 68.76 s (r:0, l:None, u:None)
P: inv: 4.62 s (r:0, l:None, u:None)
[       OK ] (3/3) EESSI_NumPy %scale=1_node %module_name=SciPy-bundle/2023.11-gfbf-2023b /ae66ca75 @snellius:rome+default
P: dot: 3.09 s (r:0, l:None, u:None)
P: svd: 48.43 s (r:0, l:None, u:None)
P: cholesky: 5.67 s (r:0, l:None, u:None)
P: eigendec: 94.73 s (r:0, l:None, u:None)
P: inv: 4.8 s (r:0, l:None, u:None)

Interestingly enough, eigendec seems to be substantially faster for 2023a and b, but no difference for 2022b. Whatever the reason, performance with thread binding is the same or better, so... I think this is good to merge :)