Update profiling setup

[max-models]

Updated scope-profiler to 0.3.3. There are some big improvements in the postprocessing in this version. Especially for GPU profiling and also line-by-line profiling for the regions we already defined using the tool without having to use line_profiler directly.

For example, please try running this param file:

from struphy import ProfilingOptions, Simulation
from struphy.models import Poisson

model = Poisson()

profiling_opts = ProfilingOptions(use_line_profiler=True)

sim = Simulation(
    model=model,
    profiling_opts=profiling_opts,
)

sim.run(profiling_activated=True)

Then inspect the profiling file in the terminal with

scope-profiler tui sim_1/profiling_data.h5

With scope-profiler > 0.3.0, there is a new optional syntax for the profiling manager. Instead of writing ProfilingManager.setup(...) and then ProfilingManager.finalize(), it is now possible to define the start and stop of the profiling manager work using contexts, just as is done with the profiling regions.

For example (from the scope-profiler readme

from scope_profiler import ProfileManager

with ProfileManager.session():
    @ProfileManager.profile("main")
    def main():
        with ProfileManager.profile_region("work"):
            sum(range(100))  # replace with the code you want to measure

    main()
# writes profiling_data.h5 and prints a summary

For us, this means we can use this syntax in sim.run(), so we don't need the _setup_profiling() method. Everything in sim.run() is just indented by one tab after with ProfileManager.session(...):, but in the diff it looks like there is a much bigger diff... I promise there isn't though.

Changes in the code:

1. Added functions=[...] to ProfileManager.profile_region calls which are profiling a very small scope, most likely just a function. When line profiling is turned on, this means that the lines inside that function are profiled, not just the line for the function call itself.

For example:

with ProfileManager.profile_region(self._solve_region, functions=[self._A_inv.solve]):
            _u = self._A_inv.solve(rhs, out=self._u_new)

2. Added a ProfilingOptions class to specify profiling option in the parameter files, this class is automatically added to the scripts with struphy params MODEL. This class basically mirrors the parameters used in scope-profiler
3. Deprecated profiling_activated from the EnvironmentOptions class and moved it to sim.run(profiling_activated=profiling_activated) instead
4. Added tutorials/tutorial_scope_profiling.ipynb
5. Updated docs
6. Updated dependencies for gvec, vtk, feectools. I also set lower bounds for feectools and scope-profiler.

[github-actions]

📖 Docs preview: https://struphy-hub.github.io/docs-preview/pr-354/

[spossann]

That looks great - looking forward to testing it!