Error checking for DPStokes open

[rykerfish]

This is for the error checking for DPStokes open that we talked about. I'm leaving it as a draft currently since I'm still unsure about the tolerances.

Madison has a case for bending forces on a membrane where I recall the absolute sum being ~1e5 but the (signed) sum being ~1e-2 or 1e-3. This is kinda small and seems like it should be acceptable, but overall not that small numerically. I'm asking for the exact values to decide.

@RaulPPelaez @brennansprinkle how do you think we should set the tolerance for when DPStokes throws an error? The most sound way to be seems to be setting it relative to the absolute mean, i.e.

$$\frac{\sum F_i}{\sum \vert F_i \vert} < \epsilon $$,

where $\epsilon$ could seemingly be set pretty small, maybe at least $10^{-4}$, based on Madison's case. However, I think this would require an extra reduction per call to Mdot which might get expensive. We could alternatively set it conservatively based on the signed sum, and just informing users with cases like this to enable the unsafe flag.

Closes #78

[github-actions]

✅ Linter reported no issues

All Python files are correctly formatted with Black.

[github-actions]

✅ Linter reported no issues

All C/C++ files are correctly formatted with clang-format.

[RaulPPelaez]

the absolute sum being ~1e5 but the (signed) sum being ~1e-2 or 1e-3

Is that computing the sum with a parallel reduction or just sequential summing? The parallel version might be more precise. A quick and dirty check would be to accumulate using double (you can use thrust::transform_reduce).
Otherwise your strategy sounds reasonable to me.

You can compute the absolute sum by using a single transform_reduce by transforming the real3 forces into a vec4 (fx,fy,fz, abs(fx)+abs(fy)+abs(fz)) and accumulating that.

[RaulPPelaez]

LGTM, feel free to merge