Add dois/doi_messages tables and data-access methods

[baltierra]

Adds two tables to support tracking which messages have been used to mint DOIs, enabling duplicate-DOI detection for consumers like Hermes without depending on Zenodo's API for that lookup.

archive/migrations/0006_add_doi_tracking.sql

• dois: one row per minted DOI (doi string, record URL, title, creator, timestamp). Unique index on doi.
• doi_messages: many-to-many between DOIs and message UUIDs. Composite primary key (doi_id, message_uuid), cascade-delete FK to dois, and an index on message_uuid for fast reverse lookup ("which DOIs is this message part of?").

No FK to messages.uuid — that column only has a non-unique index, not a unique constraint, so a hard FK could break inserts in edge cases already handled elsewhere (duplicate message ingestion).

archive/database_api.py

• sqlalchemy.Table declarations for both new tables, following the existing pattern used for messages/topics.
• insert_doi(doi, record_url, title, created_by, message_uuids): inserts one dois row plus its associated doi_messages rows in a single transaction.
• find_dois_for_messages(message_uuids): given a candidate set of UUIDs, returns every message belonging to any DOI that overlaps at all with that set (not just the overlapping members). This lets callers distinguish an exact-duplicate package (same message set, same size) from a partial overlap (a message reused inside a different, larger package) using only counts — without needing a second round-trip query.

archive/access_api.py

• Thin pass-through wrapper methods (record_doi, find_dois_for_messages) added to Archive_access, matching the existing convention. No business logic here, consistent with every other method in this class.