Inverse modelling (adjoint) improvements

[cpjordan]

Another list of improvements:

• Add notes on how to reduce memory consumption for big problems (PETSc garbage collector goes into reduced_functional.py of pyadjoint - this doesn't seem to work when only placing this in the inversion tools)
• Add TAO/ROL/IPOPT optimisers
• Add simultaneous optimisation of elevation and velocity (or any combination of fields, e.g. turbine power too if we have that)

I've already started the first two, but I need PR #371 approved first.

[cpjordan]

Also - look at schedulers as this might remove the need to manually modify pyadjoint

[cpjordan]

* [ ] Add notes on how to reduce memory consumption for big problems (PETSc garbage collector goes into reduced_functional.py of pyadjoint - this doesn't seem to work when only placing this in the inversion tools)

See #430. Scheduler addresses this issue.

* [ ]  Add TAO/ROL/IPOPT optimisers

See new-adjoint-optimisers for ROL, but had also tested TAO. It doesn't seem worth the additional code as I've not seen any benefit in memory, parallelisation etc. to optimising with anything other than SciPy.

* [] Add simultaneous optimisation of elevation and velocity (or any combination of fields, e.g. turbine power too if we have that)

See #429