Return an explicit empty Memory recall reason

[hamzamerzic]

Summary

• add the stable no_relevant_result reason to structured receipts whose status is empty
• keep empty lookup as a successful outcome and preserve the existing human-readable answer
• make the Python test suite runnable directly without a caller-supplied import path
• bump the app version to 2.2.3

Context

This is a compatible follow-up to #14, which introduced structured hit/empty/failed outcomes, and #23, which added safe failure reasons. Unlike failure reasons, no_relevant_result describes a successful lookup that selected no notes; older consumers continue to rely on status: empty and may ignore the additive field.

Validation

• pytest -q --tb=short (83 passed)