Declare conversion assignments for Tpetra vectors.

[bangerth]

As noted in #16636 (comment) by @jpthiele , we need conversion assignment operators for Tpetra vectors. I had actually started to write those, but don't have time today or tomorrow to finish them. I'll upload it in case someone else wants to finish my work -- this compiles for several more tests, but obviously produces linker errors.

[jpthiele]

This is certainly a good step but it also would be interesting if we could manage shallow copies
between the types so we can use a Tpetra solver and the built-in solution vector updates.

But for that we possibly need all that view stuff and @masterleinad might now how to manage that.

[bangerth]

That raises difficult ownership questions. My suggestion would be to get these operators implemented first. We can implement more elaborate schemes later on.

[jpthiele]

That raises difficult ownership questions. My suggestion would be to get these operators implemented first. We can implement more elaborate schemes later on.

I totally agree that that is the more elaborate one and should be thought about carefully.
And we don't necessarily need to expose that as a functionality to the user.
But it would be much more convenient for the user to pass Vector or LA::d::Vector to the solve functions in SolverDirect and the iterative counterpart.

[masterleinad]

Superseded by #16663.