PERF: speed up the adjustment of SNR

[ctrltz]

The SNR adjustment is currently very slow (13 seconds) when the number of noise sources is 500 (see detailed reports below):

• 9 seconds are spent for 499 calls of _combine_stcs. In this PR, _combine_sources_into_stc is reworked to combine all sources at once instead of adding them one by one.
• 3.77 seconds are spent for 6 calls of _get_sensor_space_variance (1 for all noise + 5 for alpha sources). While the calculation of covariance should be the correct approach to account for correlations between sources, in practice it might be computationally easier to actually project the data to sensor space and average the squared values to get mean variance. This strategy is also implemented in this PR.

Current version on master:

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
        1    0.000    0.000   13.749   13.749 /home/willenjoy/Projects/roiextract-simulations/scripts/misc/profile_meegsim.py:11(simulation_ssd_pattern_recovery)
        1    0.000    0.000   13.694   13.694 /home/willenjoy/Projects/meegsim/src/meegsim/simulate.py:187(simulate)
        1    0.001    0.001   13.694   13.694 /home/willenjoy/Projects/meegsim/src/meegsim/simulate.py:245(_simulate)
        1    0.001    0.001   13.297   13.297 /home/willenjoy/Projects/meegsim/src/meegsim/snr.py:104(_adjust_snr)
        1    0.095    0.095    9.526    9.526 /home/willenjoy/Projects/meegsim/src/meegsim/sources.py:173(_combine_sources_into_stc)
      499    0.016    0.000    9.349    0.019 /home/willenjoy/Projects/meegsim/src/meegsim/utils.py:12(combine_stcs)
     1497    4.617    0.003    4.631    0.003 /home/willenjoy/miniforge3/envs/roiextract-simulations/lib/python3.12/site-packages/numpy/lib/function_base.py:5368(insert)
12917/505    0.017    0.000    4.551    0.009 /home/willenjoy/miniforge3/envs/roiextract-simulations/lib/python3.12/copy.py:118(deepcopy)
  529/505    0.004    0.000    4.545    0.009 /home/willenjoy/miniforge3/envs/roiextract-simulations/lib/python3.12/copy.py:247(_reconstruct)
      535    0.004    0.000    4.532    0.008 /home/willenjoy/miniforge3/envs/roiextract-simulations/lib/python3.12/copy.py:217(_deepcopy_dict)
      499    0.001    0.000    4.524    0.009 /home/willenjoy/miniforge3/envs/roiextract-simulations/lib/python3.12/site-packages/mne/source_estimate.py:1099(copy)
     2128    4.502    0.002    4.502    0.002 {method '__deepcopy__' of 'numpy.ndarray' objects}
        6    3.526    0.588    3.770    0.628 /home/willenjoy/Projects/meegsim/src/meegsim/snr.py:9(get_sensor_space_variance)

The same simulation after the optimization:

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
        1    0.000    0.000    1.195    1.195 /home/willenjoy/Projects/roiextract-simulations/scripts/misc/profile_meegsim.py:11(simulation_ssd_pattern_recovery)
        1    0.000    0.000    1.157    1.157 /home/willenjoy/Projects/meegsim/src/meegsim/simulate.py:190(simulate)
        1    0.007    0.007    1.157    1.157 /home/willenjoy/Projects/meegsim/src/meegsim/simulate.py:255(_simulate)
        1    0.008    0.008    0.608    0.608 /home/willenjoy/Projects/meegsim/src/meegsim/snr.py:114(_adjust_snr)
        2    0.000    0.000    0.542    0.271 /home/willenjoy/Projects/meegsim/src/meegsim/source_groups.py:79(simulate)
        2    0.001    0.000    0.542    0.271 /home/willenjoy/Projects/meegsim/src/meegsim/sources.py:130(create)
        6    0.012    0.002    0.518    0.086 /home/willenjoy/Projects/meegsim/src/meegsim/snr.py:10(get_sensor_space_variance)
        2    0.001    0.001    0.514    0.257 /home/willenjoy/Projects/meegsim/src/meegsim/waveform.py:68(one_over_f_noise)

[ctrltz]

Tried to change too many things at once, will split in two PRs.