fix: Lock while generating the Security Key

[febinkdominic]

Fixes duplicate key generation when a scoped JwtService rotates keys concurrently across requests #65

• Added a process-wide SemaphoreSlim RotationLock in JwtService so concurrent requests rotate the current key once, not once per request.
• GetCurrentSecurityKey now double-checks under the lock before revoking + generating.

Fixes duplicate key generation when a scoped JwtService rotates keys concurrently across replicas

• Issue: When run on multiple pods (Kubernetes), the pods independently generate new keys and cache the last two keys. On a cluster with 3 replicas with AlgorithmsToKeep = 2, all three generate new keys and cache different pairs of keys.

• Fix: Use the previous valid key's Id, and increment it, and use that to generate the GUID of the new key - Every instance generates the same GUID and tries to insert, the database fails one of the inserts acting as the coordinator.

• Cross-replica coordination only works on EF Core store - Database is used as the coordinator when multiple replicas race.

• EF Core store: computes a deterministic Id from SHA256(use:kty:version) so concurrent inserts collide on the primary key. On collision/transient fault it detaches its own entity, re-reads the winner, and returns it (throws only if no winner exists).

• InMemoryStore, DataProtectionStore, FileSystemStore updated to the new signatures.

• Changed caching from Sliding to Absolute. When a key is revoked, the key gets refreshed after the cache expires instead of renewing the stale cache.