Destroy Attributes's handle before its parent_link

[FlyingSamson]

As pointed out by @jkotan in a comment to the related PR #660, the same issue as described in #659 also appears for Attributes:
If an Attribute instance is kept around after the last reference to a file was dropped in client code, the order of Attribute's members causes it to first try to close the underlying file before closing its own handle. If the file was opened with close degree Semi or the MPI-IO file driver was used, closing the file while the Attribute is still open is prohibited.

This PR, therefore, applies the same fix to Attribute as was applied to Node in #660; that is, it reorders the member of Attribute to ensure that the handle is destroyed first, before attempting to close the file through destroying the link.

[ggoneiESS]

lgtm, same as fix for #660 as described in PR

[jkotan]

I was just thinking, since it is hard to add proper unit tests maybe it would be good to add a small comment in the code about the issue in order no one revert it back.

[ggoneiESS]

I was just thinking, since it is hard to add proper unit tests maybe it would be good to add a small comment in the code about the issue in order no one revert it back.

I don't see the harm in adding a comment to make it less opaque, but I plan to see if a test can be created using rdbuf, something like:

    std::ostringstream outputForTesting;
    auto* originalCerrBuffer = std::cerr.rdbuf(outputForTesting.rdbuf());

    // catch runtime_error

    REQUIRE(outputForTesting.str() == "Failed ~ObjectHandle:\n");

[jkotan]

I merge the current PR. We can add possible tests in a new PR