Prime reranking auth before parallel batches

[erichare]

What this PR does:

Runs reranking batch 0 to completion before constructing and subscribing to the remaining batches. A successful first batch primes the upstream API Gateway authorizer cache; the tail batches then retain their existing concurrent, fail-fast execution.

This reduces the authorization cache-miss fan-out of one multi-batch findAndRerank request from as many as one miss per batch to one miss per request. Concurrent top-level requests can still race each other, so this is intentionally paired with the Lambda timeout and resilience changes in riptano/cloud-helm-charts#1095.

The tradeoff is one reranking-batch round trip on the critical path before the concurrent tail starts.

Which issue(s) this PR fixes:

Companion to riptano/cloud-helm-charts#1095.

Validation:

• JAVA_HOME=/Library/Java/JavaVirtualMachines/graalvm-21.jdk/Contents/Home ./mvnw -B -ntp -Dstyle.color=always clean test (3,424 tests)
• Focused reranking provider and gateway tests (10 tests)
• Added coverage for first-batch sequencing, concurrent tail execution, first-batch failure, single-batch requests, empty passages, result indexes, and model usage aggregation

Checklist

• Changes manually tested in a deployed environment
• Automated Tests added/updated
• Documentation not required (no API or configuration change)
• CLA Signed: DataStax CLA

[github-actions]

Unit Test Coverage Report

Overall Project	53%	🍏
Files changed	100%	🍏

File	Coverage
RerankingProvider.java	59.34%	🍏

[github-actions]

Integration Test Coverage Report (dse69-it)

Overall Project	71.39% -0.08%	🍏
Files changed	0%	❌

File	Coverage
RerankingProvider.java	12.09% -18.9%	❌

[github-actions]

Integration Test Coverage Report (hcd-it)

Overall Project	72.71% -0.08%	🍏
Files changed	0%	❌

File	Coverage
RerankingProvider.java	12.09% -18.9%	❌

[amorton]

-1 blocking until we know more , at the surface level this does not feel like the best idea, we should fix the upstream problem