⚡️(auth) cache token introspection and service resolution #184

Problem

Each search/delete request:

1. calls the OIDC provider's introspection endpoint (ResourceServerAuthentication, no caching), and
2. queries Postgres for the Service matching the token audience, plus its sub-services (indexing.py#L84-L91).

Each index request also queries Service by token, and runs an indices.get on OpenSearch.

This adds latency to every search (search-as-you-type makes it worse) and puts load on the identity provider. It also makes Find unavailable whenever the provider is slow or down.

Proposal

• Cache introspection results per token hash for a short TTL, bounded by the token's exp (check whether django-lasuite supports this already).
• Cache the Service → allowed-indices resolution, invalidated when a Service is saved.
• Replace LocMemCache with a shared cache (Redis is already deployed) so the cache works across replicas.