feat(e2e): Crossplane DLQ support + drift detection tests + cold-start scaling

[atemate]

Summary

• SQS DLQ queue: Add shared asya-{namespace}-dlq Crossplane Queue resource in asya-crew chart. Each actor queue gets a redrivePolicy pointing at it (constructed from awsAccountId + awsRegion in chart values).
• Un-skip DLQ tests: Remove @pytest.mark.skip from test_poison_message_moves_to_dlq_e2e and test_dlq_preserves_message_metadata_e2e.
• Crossplane drift tests: Rewrite 4 queue-health tests — replace "operator recreates queue" framing with Crossplane drift reconciliation, reduce timeout from 360s to CROSSPLANE_RECONCILE_TIMEOUT_SECONDS (default 120s).
• Cold-start test: Add test_cold_start_backlog_processing — scales actor to 0, enqueues 20-message backlog, asserts 90%+ complete after KEDA scales up.

Merges aints: 1f3k (DLQ un-xfail) + 1fbq (queue health un-skip) + 1f2y (cold-start scaling) → single task 1f2y.

Test plan

• helm lint deploy/helm-charts/asya-crew/ passes
• helm lint deploy/helm-charts/asya-crossplane/ passes
• helm template with awsAccountId=000000000000 shows redrivePolicy in SQS queue
• helm template with awsAccountId= shows no redrivePolicy
• DLQ tests collected without skip (grep h6h2 returns 0)
• Queue health tests collected without skip (grep pwx6 returns 0), renamed to test_crossplane_recreates_*
• Cold-start test collected, polls for actual pod drain before enqueuing
• make lint clean

[gemini-code-assist]

Code Review

This pull request effectively adds SQS DLQ support, enables the corresponding tests, and refactors the queue health tests to align with the new Crossplane-based reconciliation strategy. A new cold-start scaling test is also introduced, which is a great addition for ensuring scale-from-zero reliability. However, a security audit identified two injection vulnerabilities in the Crossplane composition template where Helm values are rendered without proper escaping. Specifically, a potential template injection via awsAccountId and a JSON injection via maxReceiveCount were found. These issues should be remediated by using the toJson filter to ensure values are safely handled during template rendering. Additionally, the implementation of the new cold-start test could be optimized, as waiting for task completions can be significantly improved to reduce test execution time.