Always export final solution

[ddundo]

Closes #106

With this PR the final solution of each subinterval always gets exported, instead of the first one.

Also tagging @stephankramer and @acse-ej321 for awareness :)

[jwallwork23]

Thanks for this @ddundo.

Please could you use reversed instead of np.flip? Good to use Python build-ins rather than external packages, where possible.

Could you also do the same thing in adjoint.py?

[ddundo]

Ah sorry @jwallwork23, tunnel vision here! I was looking into adjoint.py now and I'm not sure I understand Line 369...

Shouldn't it be project(sols.adjoint[i + 1][0], sols.adjoint_next[i][j]) instead of project(sols.adjoint_next[i + 1][0], sols.adjoint_next[i][j])? That is, the adjoint_next on the final subinterval timestep is the adjoint on the first timestep in the next subinterval?

[ddundo]

This should be ready for a re-review now @jwallwork23 :)

I also made the change I mention in the above comment. It looked like a bug to me

[jwallwork23]

Thanks very much for this @ddundo! Your bugfix and test tweaks make sense to me.

Just two very minor requests. If you fix those then feel free to merge.

[jwallwork23]

One final thing for conciseness: could we switch this to

for j, block in enumerate(reversed(solve_blocks[::-stride])):

so that it (hopefully) fits on one line? My attention was brought to this because I was concerned that zip doesn't care if the iterables have different lengths so might be skipping some iterations, but the check above ensures things are okay.

[ddundo]

Nice! Done in eb187c0

[jwallwork23]

Similarly here.

[ddundo]

Done in eb187c0

[ddundo]

Thanks Joe! I did as you suggested and will merge :)