Fix point location in parallel

[connorjward]

Description

I believe that this fixes #3151 and supersedes #3293. It certainly seems to fix this issue on my machine.

[github-actions[bot]]

	Tests	Passed ✅	Skipped ⏭️	Failed ❌
Firedrake complex	8078 ran	6459 passed	1582 skipped	37 failed

[github-actions[bot]]

	Tests	Passed ✅	Skipped ⏭️	Failed ❌
Firedrake real	8084 ran	7262 passed	786 skipped	36 failed

[connorjward]

Turns out that this is harder than expected. With my changes, if we consider the interval mesh

           p0   |      p1
        o-----o | -----o-----o
                |

and want to locate points A and B on p0 (note the extra cell from p1 is not visible)

           p0   |   p1
        o-----o | -----o
                |  A     B

then our strategy for identifying missing points means that both A and B are considered as being claimed by 'cells that we can't see' and both are removed. This is clearly wrong.

Unfortunately I can't currently think of a way to fix this in a non-invasive manner at present. I do have some general thoughts:

• Point location should probably not be being done for ghost cells. IMO those should be handled by the other process and communicated over.
• We should look to use star forests more as the existing approach, with each process seeing all points, is not very scalable.

[connorjward]

Closing as this was poorly thought out (though in my defence this is extremely confusing). I will attempt to get #3293 merged.