Implement piecewise-linear rhod profile in condensation methods

[mikhailmints]

Closes #746, closes #765.
This implements a piecewise-linear rhod profile in the condensation methods instead of using rhod_mean. This change makes the condensation solver much less dependent on the timestep in the parcel environment, particularly for parcels with low velocities. For example, before the change, a particular run of the parcel model resulted in the following:

There was a very significant difference between the predicted supersaturation for different timestep values, which caused the solver to significantly overestimate the peak supersaturation and activation fraction unless the timestep was set to a very low value.
After the change, the results for different timesteps become nearly identical:

TODO: implement the same changes in the ThrustRTC backend.

[slayoo]

Thank you, @mikhailmints !

[slayoo]

Also, the plot you included seems a ready-to-use base for a new unit test! Asserting that the two lines match.
In many unit tests, we include a plot=False argument, which if turned to True, makes the test produce a plot (just before an assert) - so the plot can be included as well, just adding at the end:

        if plot:
            pyplot.show()
        else:
            pyplot.clf()

[claresinger]

Thank you @mikhailmints - looking great! Can you check for convergence with this new linear rhod profile with dz steps rather than dt steps? (Using dz rather than dt will remove the w dependence.) Maybe ranging from dz=0.01 to dz=10m? That would be helpful to know for the examples. You could add this as a plot/convergence test as a unit test, as Sylwester suggested above.

[trontrytel]

Looks like macos-12, 3.7, smoke_tests/parcel and macos-12, 3.10, smoke_tests/parcel are timing out. Not sure what to do about them - could we split that chunk of tests into two to deal with the timing?

I'm bumping this PR, because I think it's a really beneficial update to the default behavior ;)

[slayoo]

@trontrytel @mikhailmints @claresinger

Concerning the timeouts on macOS (which occur within SciPy code, not the PySDM solver), I'd say let's just increase the timeout.
However, I vote for waiting with merging till we have a test added that checks what's visible in the above plots - to make sure that the behaviour is asserted on CI, and that it will be checked for in future developments. @mikhailmints could you give a try at adding the new smoke test?

We have also the TODO of matching the changes in the ThrustRTC backend, but if well labelled and marked with a new issue, we can defer it for later.

Thanks!

[slayoo]

The README parcel example is also a good test case here. If we run it for double the ascent altitude, there is some noise present, which disappears with the changes in this PR. Aiming at committing a README-based test case here soon.

[slayoo]

closing and reopening to fix commit list