Improvements to bubble shear demo

[ddundo]

• [x] Vast majority of effort in the bubble shear demo goes into interpolating velocity u.interpolate(velocity_expression(mesh, t))
  • The reason is that t in velocity_expression is hardcoded as a float so a lot of time is spent recompiling it. Instead, it should be wrapped in Constant (or Function of R). This made it much faster.
  • Doing u.dat.data[:] = u_numpy, where u_numpy is the velocity computed with numpy also works and is actually faster than the above, but it's not very firedrake-y, so I went with the above method
• [x] The first computed metric in metric advection should be normalised I think
• [x] Wrong description in Line 416
• [x] Mention explicit timestepping and CFL condition in the exercise rubric
• [ ] Recompute initial condition on the first adapted mesh, rather than projecting the one computed on the uniform mesh. This may actually give a slightly higher error in the end - put it in the exercise rubric?

[jwallwork23]

Thanks for looking into this! I also found there's plenty of optimisation to be done across the repos.