Fix handling of SkyCoords with N-d frame attributes - #5891

The failure here looks unrelated to my changes, but something to do with remote sites. Perhaps someone with write access can restart that one failing test?

@StuartLittlefair - So these are from those bugs again that are fixed in master?

To answer my question, I see both issues on master, too. I'll have a look around whether it's possible to change the base branch of this PR, if not then this should be closed and a new one opened against master.

Hmm, changing base branch clearly worked, but introduced a lot of unwanted stuff here. I'm going ahead and try to rebase this mess and force push it back to this branch. (@StuartLittlefair - I've made a backup copy of your original as bsipocz/StuartLittlefair-patch-2-copy)

Duh, I should have seen that you had a PR. I'll close mine and instead make a PR to your branch (since I think my solution is a bit more elegant...)

So, I added 2 commits for the representation. Am still trying to wrap my head around the rest of the PR (might actually be better to keep them separate, sigh...).

I think the solution is not right -- transform_to is the problem, in that it doesn't do what it states it does - it should remove attributes before initializing a new class.

@mhvk, I only really bundled these together because I wasn't smart enough to find a simple regression test for #5889.

Perhaps the simplest way forward is for me to close this, for you to re-open your PR for #5889 and we'll look again at a separate PR for #5890, which probably does need some careful thought about what exactly transform_to should be doing?

See #5897 for a simpler fix, which combines all PRs.

OK, it seems like for now this PR probably should be closed (we can reopen if #5897 turns out to be problematic as well).