Fix duplicates when mixing explicit and autogenerated ids

[cole]

When some metadata section ids, but not all, are autogenerated, duplicates can easily occur. This PR changes metsrw to track ids used, and if a conflicting id is generated, retry.

I'm not totally happy with this implementation, I think if we had a more consistent metadata class structure, we could easily do this in a base class, but it's an improvement on what was there.

I've also made id_string a property (vs a method) in
bdf84b6, which is a small API change that will require updates to a few lines in Archivematica (all in archivematicaCreateMETSReingest.py). That change can be dropped if it is problematic, it just seemed like an easy improvement.

Connects to archivematica/Issues#442.

[cole]

Works in my testing when combined with the changes in artefactual/archivematica#1407

[sevein]

I've tested this and it works. Thanks @cole!

Just for the record, it's still true (as it was in v0.3.3 when we removed random IDs) that the ID generators are going to be shared across multiple documents. E.g. the following test passes

def test_idgenerator_instance_attribute(self):
    doc_foo = metsrw.METSDocument()
    fsentry = metsrw.FSEntry("foo.txt", file_uuid=str(uuid.uuid4()))
    fsentry.add_premis_object("<premis>object uno</premis>")
    fsentry.add_premis_event("<premis>event uno</premis>")
    fsentry.add_premis_event("<premis>event dos</premis>")
    doc_foo.append_file(fsentry)

    doc_foo = doc_foo.serialize()
    assert doc_foo.find(".//mets:amdSec", namespaces=metsrw.utils.NAMESPACES).get("ID") == "amdSec_1"
    assert doc_foo.find(".//mets:techMD", namespaces=metsrw.utils.NAMESPACES).get("ID") == "techMD_1"
    assert doc_foo.findall(".//mets:digiprovMD", namespaces=metsrw.utils.NAMESPACES)[0].get("ID") == "digiprovMD_1"
    assert doc_foo.findall(".//mets:digiprovMD", namespaces=metsrw.utils.NAMESPACES)[1].get("ID") == "digiprovMD_2"

    doc_bar = metsrw.METSDocument()
    fsentry = metsrw.FSEntry("bar.txt", file_uuid=str(uuid.uuid4()))
    fsentry.add_premis_object("<premis>object uno</premis>")
    fsentry.add_premis_event("<premis>event uno</premis>")
    fsentry.add_premis_event("<premis>event dos</premis>")
    doc_bar.append_file(fsentry)
    print(doc_bar.tostring(encoding="unicode"))

    doc_bar = doc_bar.serialize()
    assert doc_bar.find(".//mets:amdSec", namespaces=metsrw.utils.NAMESPACES).get("ID") == "amdSec_2"
    assert doc_bar.find(".//mets:techMD", namespaces=metsrw.utils.NAMESPACES).get("ID") == "techMD_2"
    assert doc_bar.findall(".//mets:digiprovMD", namespaces=metsrw.utils.NAMESPACES)[0].get("ID") == "digiprovMD_3"
    assert doc_bar.findall(".//mets:digiprovMD", namespaces=metsrw.utils.NAMESPACES)[1].get("ID") == "digiprovMD_4"

That's not an impediment for Archivematica in the way that it's currently used though. Also users hitting this limitation can clear the generators like @cole is doing in test_metadata.py.