ops: revise kokkos matrix-vector product

[mzuzek]

refs #446

[fnrizzi]

I find the use of impl::execution_space and has_execution_space confusing. Seems to me that you are defining one to aid the other. I think we should just have has_execution_space and specialize it for kokkos tpetra and tpetrablock. This would make things easier.

Please make a separate PR for this execution space trait thing.

[mzuzek]

@fnrizzi As we have discussed and planned, I rebased this PR on #502 which removes have_matching_execution_space predicate all together. This obsoletes c40b2c473698dc87184d7eaccfcd8b36ae0a62cf (making it empty) and hopefully resolves all raised issues and questions on execution_space and has_execution_space utilities which supported that predicate.

[mzuzek]

for the record, we had weird CI failure that did not reproduce on the re-run:

terminate called after throwing an instance of 'std::invalid_argument'
  what():  stoi

[mzuzek]

@fnrizzi

Kokkos test coverage

Both nontranspose and transpose overloads of pressio::ops::product() are tested with all combinations of plain views and available Pressio expressions:	A	x	y
view2d	view1d	view1d
view2d	span	view1d
view2d	diag	view1d
view2d	view1d	span
view2d	view1d	diag
view2d	span	span
view2d	span	diag
view2d	diag	span
view2d	diag	diag
subspan	view1d	view1d
subspan	span	view1d
subspan	diag	view1d
subspan	view1d	span
subspan	view1d	diag
subspan	span	span
subspan	span	diag
subspan	diag	span
subspan	diag	diag

Note: A could be also pressio::as_diagonal_matrix() in future - once it's implemented, see #304

[fnrizzi]

@mzuzek

• the table is awesome !
• i only found two things, if you can please fix those i will then merge it