[moetemp]

What changed?

This PR pins Core to the in-process stream service and reaches it from Python, with records and errors. A small pyo3 module builds a store from a serialized StreamStoreConfig. scripts/gen_bridge_client.py generates a typed StreamService, so no stream call is hand-written. Lang takes the plaintext content_hash and the batch digest before the codec, and error_from_failure maps each of Core's StreamFailure kinds to one error class.

Why?

The store runtime lives in Core now, so every Core-based SDK binds one copy of it (DD-86). Only lang has the plaintext, so hashing before the codec keeps retries deduped (DD-11). One failure kind per failure means every SDK raises the same error (DD-21).

How did you test it?

Link to a test plan if any -

• Unit Tests
• Staging
• End to End Tests

Ran the bridge store tests on the dev server with Core's memory store. The digest test checks against the vector Core's tests pin.

[moetemp]

Replaced by the Brook v3 stack, #322 to #338.