lib/model, lib/protocol: Fix file comparisons (fixes #8594)

[calmh]

Currently there's a couple of problems with file comparisons introduced in the xattr+ownership work.

1. The comparisons for when xattrs and ownership differ were logically wrong. I made it so that two files are considered equal if one has nil platform data for a given platform, and the other one has data for that platform. The intention was that when a file comes from Windows and we add Linux data locally, it shouldn't cause a resync everywhere. But that's wrong, because the added Linux data might be of interest to other Linux hosts...
2. There's a comparison for "equivalence" before we insert a fileinfo in the database. If it's "equivalent" to the global version we use that instead. That meant that the inode change time we added was forgotten, because the global version generally has a zero inode change time (from the network) and this was considered equivalent to the one we just scanned. It also meant we didn't store xattr-only changes if there wasn't already xattr data present, as per number one above.

This attempts to fix that, by making the comparisons stricter. The effect will be more changes and more syncs, but the changes to apply should be minimal.

This very likely also fixes #8604 and fixes #8590.

[imsodin]

The generally stricter ownership/xattr checking is clear enough. The flag for the strict inode change time is a bit harder to reason about. Also they are independent, right? Maybe split them up in two PRs.

For the change time: The strict flag is only needed in a code branch for a receive-only optimisation, to not create an insignificant local change that will be resolved by the subsequent pull. I don't think this had any other intended purpose nor can I see one now. I'd rather move those conditions behind a receive-only flag than adding the strict inode change time flag that's only used there.

[imsodin]

In maybe a bit clearer (and fewer) words:
For the equivalence check the current behaviour seems correct for recv-only folders (and recv-enc), and it's only intended to solve aproblem for receive-only folders. Thus I propose to keep the current check but move it behind a folder.Type in [recv-only, recv-end] condition.

[calmh]

I implemented what I think you mean, in hiding the equivalence check behind a condition on the folder type, and removing the strict checking mode which is no longer required.