This correctly handles wrapped errors as well.

This makes it possible for memfile to handle mode changes, ctime
changes, in addition to the already existing file size changes.

I'm really not especially happy with this.

It works, it's minimally invasive, and it feels like the wrong solution.

I'm pretty sure that the right solution is a breaking change of the
Request structure, to just flat out contain FileAttrFlags and FileStat,
having requestFromPacket do the parsing into those structures.

But that's an API change, and I doubt that we want a major version bump
for this.

As an alternative, we could have a new Request structure type, with that
layout, and have public functions for converting between them.

That would at least make it possible for someone to construct the
Request structure for a Setstat command.

The use case of this is being able to populate a memfile backed sftp
server with test data that has specific timestamps, as part of the test
framework of an internal program.

I'm definitely open to other ideas on how to solve this problem.

This would replace the SetStat public function from the last commit.

Instead, this makes it practical to construct a Setstat request.

From some quick testing, this appears to do the job correctly, and it
feels like a lot less of hack.

It still doesn't feel entirely right, but it's closer.

This is in favor of an alternate approach, we can always pull it back
out of git history if we want to.

It will be replaced with an alternate method.

This should cover every single case where we were using == or != on an
err.

There may be other cases to address, but this covers a big one.

With us no longer offering an interface with SetStat, there is no reason
for this to be it's own function anymore.

This also cleaned up how we handle truncation errors.

For the sake of sanity, we really should be offering _some_ way for
users of memfile to get the file mode right.  These two are what we use
internally, so they make perfect sense to expose.

We now check to ensure that the permissions on a given file allow the
user to perform the requested operation.

For Filereader, the file must be readable.

For Filewriter, the file must be writable.

For OpenFile, the file must be readable and writable.

Missed this in the first pass.

FileStat.MarshalTo takes in flags, and generates the []byte that goes into
Request.Attr

FileAttrFlags.ForRequest() generates the uint32 bitmap that goes into
Request.Flags

This reverts commit b1d53f7.

This reverts commit d41c7d4.

This reverts commit 8ee6563.

This reverts commit 87fa218.

As best as I can tell, under POSIX chmod will follow symlinks.

However I am definitely not sure if that is the case for common sftp
servers on POSIX filesystems, we should consider checking...  Some day.

This avoids logic duplication.

That is, 0oNNN instead of 0oNNNN, this is because we don't have any
current cases where we're using the 4th digit, and so things turned into
0o0200.