feat: add extensible usage reporting

[heaoxiang-ai]

Description

Introduces an extensible usage reporting pipeline for OpenViking. The core defines reporter, extractor, sink, event, and context contracts; deployments can load an external sink class without importing Kafka or another vendor SDK into the open-source package.

The first extractor recognizes official experience-skill tool parts and emits memory recall/injection usage events.

Related Issue

N/A

Type of Change

• New feature
• Bug fix
• Breaking change
• Documentation update
• Tests

Changes Made

• Add UsageReporter, UsageExtractor, and UsageSink abstractions.
• Add configurable custom-class loading for deployment-provided sinks.
• Parse completed search_experience and read_experience tool parts.
• Dispatch extracted events after session commit using the hydrated message view.
• Add the official ov-experience-memory skill contract and architecture documentation.
• Add unit and integration coverage for event extraction and reporter behavior.

Testing

• tests/unit/usage_reporter: 5 passed.
• Ruff check, Ruff format check, and git diff --check passed.
• The session integration test could not initialize locally because the installed native wheel does not expose PersistStore; the test body was not reached.

Checklist

• Code follows the project style.
• Tests cover extractor and reporter behavior.
• Kafka/vendor dependencies are not imported by the OpenViking core.
• Changes are independent from trajectory snapshots and user settings.

Screenshots

N/A

Additional Notes

Commercial or private deployments can install a separate Python package and configure its sink class through the OpenViking server configuration.