Add more checks to DirectIODirectory? [LUCENE-9788]

Followup to #10813 (which just added an assumption to the tests so they will pass on unsupported filesystems).

If you try to use DirectIODirectory on an unsupported FS like tmpfs (I am sure there are others), you'll get some nasty behavior: EINVAL from the kernel (translates to a FilesystemException), leftover files on the filesystem, etc. See LUCENE-9774's PR for some more explanation.

Maybe we could try to make the behavior friendlier/clearer in some way. It will behave like a bug today.

Some options might be:
- improve the javadocs / make javadocs stand out better.
- catch the evil EINVAL from this specific place and add a suppressed "explanation" that helps (I think we've done similar stuff before with the confusing OutOfMemoryError from FileChannel.map)
- probe that the thing can work against the filesystem in DirectIODirectory code itself somehow? This seems a bit scary as it means we have to write files, try to delete the leaked files when it fails, avoid masking other actual errors, etc.
- others...

---
Migrated from [LUCENE-9788](https://issues.apache.org/jira/browse/LUCENE-9788) by Robert Muir (@rmuir)

Start with DirectIODirectory and the changes from #10813, then read LUCENE-9774's PR for the reported EINVAL and leftover-file behavior. The issue needs a chosen approach before implementation; done should make unsupported filesystems fail with clearer behavior without masking other errors or leaving files behind.

Written by the indexing model from the issue text.