FileStorage/py2 uses `FS21` magic in file header, whereas
FileStorage/py3 uses `FS30` magic:

    https://github.com/zopefoundation/ZODB/blob/0e72b8b13657/src/ZODB/_compat.py#L39
    https://github.com/zopefoundation/ZODB/blob/0e72b8b13657/src/ZODB/_compat.py#L74

And if, upon opening the database, file magic does not match to what ZODB
expects, open is rejected:

    https://github.com/zopefoundation/ZODB/blob/0e72b8b13657/src/ZODB/FileStorage/FileStorage.py#L88
    https://github.com/zopefoundation/ZODB/blob/0e72b8b13657/src/ZODB/FileStorage/FileStorage.py#L1625-L1630

This is done with the idea for a database, that was written from
Python2, to be rejected to be opened from Python3 and vice-versa because
strings/bytes semantics changed in between py23.

As the result, many zodbtools tests currently fail on py3 when they try
to access prepared FileStorage database in testdata, because that
database was originally prepared on py2. Here is, for example, how
test_zodbdump fails:

    ___________________________ test_zodbdump[zext-raw] ____________________________

    zext = <function zext.<locals>._ at 0x7f28530bf9d0>, pretty = 'raw'

        @mark.parametrize('pretty', ('raw', 'zpickledis'))
        def test_zodbdump(zext, pretty):
            tdir  = dirname(__file__)
            zkind = '_!zext' if zext.disabled else ''
    >       stor  = FileStorage('%s/testdata/1%s.fs' % (tdir, zkind), read_only=True)

    zodbtools/test/test_dump.py:41:
    _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
    ../ZODB/src/ZODB/FileStorage/FileStorage.py:315: in __init__
        self._pos, self._oid, tid = read_index(
    _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

    file = <_io.BufferedReader name='/home/kirr/src/wendelin/z/zodbtools/zodbtools/test/testdata/1.fs'>
    name = '/home/kirr/src/wendelin/z/zodbtools/zodbtools/test/testdata/1.fs'
    index = <ZODB.fsIndex.fsIndex object at 0x7f2852fee2b0>, tindex = {}
    stop = b'\xff\xff\xff\xff\xff\xff\xff\xff'
    ltid = b'\x00\x00\x00\x00\x00\x00\x00\x00', start = 4
    maxoid = b'\x00\x00\x00\x00\x00\x00\x00\x00', recover = 0, read_only = True

        def read_index(file, name, index, tindex, stop=b'\377'*8,
                       ltid=z64, start=4, maxoid=z64, recover=0, read_only=0):
            """Scan the file storage and update the index."""
            ...
            if file_size:
                if file_size < start:
                    raise FileStorageFormatError(file.name)
                seek(0)
                if read(4) != packed_version:
    >               raise FileStorageFormatError(name)
    E               ZODB.FileStorage.FileStorage.FileStorageFormatError: /home/kirr/src/wendelin/z/zodbtools/zodbtools/test/testdata/1.fs

    ../ZODB/src/ZODB/FileStorage/FileStorage.py:1630: FileStorageFormatError

Since zodbtools primarily work on raw data without decoding stored
pickles, unlike Zope or ERP5, it should not be a problem for zodbtools
to work on py3 with the database that was prepared on py2.

-> Adjust all tests to use FileStorage data generated on the fly based
on original files in testdata/ but with FileStorage header being
rewritten to match current python.