perf: split arc_content out of the ARC CouchDB document (P1)

[Zalfsten]

Background

Currently each ARC document stored in CouchDB embeds the full RO-Crate JSON as arc_content
inside the same document that holds metadata (metadata.events, metadata.arc_hash, etc.).

Problem

Any metadata-only query (health checks, harvest statistics, event log reads) loads the entire
RO-Crate payload — which can be several MBs — over the network unnecessarily.

Although the field-projection workaround (fields=[...] in CouchDBClient.find()) was
added as an interim fix, the root cause is the mixed document layout.

Proposed solution

Store ARC content in a separate companion document:

• arc_<id> — lightweight metadata document (hash, events, status, RDI).
• arc_<id>_content — large payload document containing arc_content.

DocumentStore.get_arc_content() fetches the _content companion; all other queries
target the metadata document only.

Acceptance criteria

• New document schema implemented and migration path documented.
• CouchDB.store_arc(), get_arc_content(), and get_metadata() updated.
• get_harvest_statistics() no longer needs a fields projection workaround.
• All existing tests updated/extended.
• No regression in API behaviour.