Reduce memory usage of GLM analysis

[rob-luke]

Running the GLM analysis is too memory hungry. This forces me to downsample more than I would like.

Discussed already in #64 nilearn/nilearn#2421 #188 #186

This is also a handy tool to help investigate this https://mne.tools/mne-nirs/auto_examples/sg_execution_times.html

[larsoner]

See also mne-tools/mne-python#8379, but also just mprof run ...; mprof plot locally can help fix things and iterate. And also memory_profiler with decorators if you want to dig into specific functions, and/or mne.utils.object_size for NumPy+builtin python type sizing for memory usage.

[G21R]

Please excuse me, I'm a beginner. I want to understand how I can apply baseline adjustment in GLM group analysis. I have seen that many parameters are used in data processing (detrending,baseline, etc) tell me how to apply it in group analysis. After all, the main parameter that is taken into account in the analysis based on GLM is the hemoglobin value (raw_haemo) how to still apply baseline for group analysis of GLM. Please forgive me for my English. And please help me please