Retry user config lookup on missing subject

[lucarlig]

Problem

Brand-new subjects can reach the dataplane before the control-plane publisher has written UserConfig:<subject> into Redis. The request currently fails immediately with a bare 400 Bad Request and Problem occurred retrieving the configuration, which hides the publisher-latency race from the client.

This showed up in live-rbac when fixtures created users and servers, then exercised them within seconds: existing-subject cache staleness is handled by the user-config cache fix, but first-publish misses for brand-new subjects still need a bounded on-miss retry.

Approach

• Retry NoDataForKey user-config lookups before rejecting the request.
• Keep the retry bounded: 30 attempts at 100ms each, for about 3 seconds max.
• Return 403 Forbidden with No dataplane config for subject if the subject still has no config after retries.
• Cover retry success and retry exhaustion in middleware tests.

Validation

• cargo test -p contextforge-gateway-rs-lib user_config_store::tests
• cargo test -p contextforge-gateway-rs-lib -- --test-threads=1
• cargo fmt --check
• cargo clippy -p contextforge-gateway-rs-lib --all-targets -- -D warnings

[lucarlig]

this is not an issue, we will modify the test instead to give time for control-plane to catch up