Summary

Fixes #11 by making no-match upserts run through the same EmbJSON processing path as normal updates.

Before this change, update(..., upsert=True) only called process_update(...) when doc_ids existed before the Mongo update. If the filter matched nothing, Mongo inserted the new document but xText / xImage fields were left raw:

• no chunks were written back into xText
• no Pinecone vector tasks were queued
• no image embedding/upload tasks were queued
• semantic query could not find the upserted content

Changes

• Add a small helper to derive the upsert document ID from _id equality filters or $setOnInsert._id.
• Generate and inject a deterministic _id into $setOnInsert when Mongo would otherwise create one implicitly.
• Use that ID to run the existing process_update(...) path before update_many(..., upsert=True) persists the document.

Validation

• python3 -m py_compile blueprints/v0/project/db/collection/document/services.py

Notes

This keeps existing matched-update behavior unchanged. The extra branch only runs when upsert=True and the pre-update filter finds no existing document.