Follow-up: refactor NotifyBackendExpired to accept metadata and use lazy expiry handling

[yrobla]

Context

Raised during review of #4404 (RC-16: Update Redis metadata when backend sessions expire).

Feedback

1. Redundant storage.Load in NotifyBackendExpired

NotifyBackendExpired currently performs a storage.Load as its first step, but there are no callers yet. Any caller that detects a backend expiry will almost certainly already have the session metadata or MultiSession in hand — otherwise it wouldn't know the backend expired.

Suggestion: Accept the metadata (or MultiSession) as a parameter to skip the redundant load.

2. Prefer lazy expiry handling over proactive health-checking ✅ Done

Backend expiry should be handled lazily (updating metadata when the session is next accessed or used) rather than via a health-checking mechanism that proactively scans for expired backends. Lazy updates are simpler and avoid the synchronization complexity of background health checkers.

Already implemented: NotifyBackendExpired does not immediately evict the cache entry. Eviction happens lazily on the next GetMultiSession call when checkSession detects metadata drift.

Proposed changes

• Refactor NotifyBackendExpired to accept session metadata or MultiSession as a parameter
• Remove the internal storage.Load call

Notes

Not blocking the original PR — this is a follow-up to be addressed when the caller is introduced.