More robust way to detect atomic move failure. Fixes #5415.

[headius]

It must have escaped my notice when improving the EXDEV error propagation that the JDK does have a specific error type for failed atomic moves. This logic should work on all OpenJDK 7+ based JDKs.

@ahorek If you can confirm this works for you locally (I have confirmed it myself on MacOS) it would be a big help!

[ahorek]

confirmed! thanks

[headius]

@ahorek @enebo I have pushed two additional commits that add more java.io and java.nio.file exceptions to this more robust logic. We can add those, or defer them if it's too much of a risk. The original commit should get cherry-picked in if that's the case.

[enebo]

I think we should just put the EXDEV fix in this alone for 9.2.1.0 and then put the rest in after release. All our other string matches have been working for years without complaint and I don't want to add another variable of uncertainty for 9.2.1.0. 9.2.2.0 can hopefully have a little bake and we can have some confidence that the rest of this is solid (which I suspect it is but I am a bit paranoid this week).

[headius]

I moved the other two commits to #5417 and merged in just the EXDEV fix for 9.2.1.