fix(ecs): preserve validated staging worker CPU

[suguanYang]

Summary

• change the staging worker ECS task size from 1 vCPU to the evidence-selected 2 vCPU while retaining 4 GiB memory and WORKER_CONCURRENCY=10
• update the ECS deployment contract so CI cannot restore the rejected 1-vCPU task size

Evidence

The one-vCPU staging topology failed the worker health gate under the measured production envelope. Two fixed 2-vCPU/4-GiB Fargate Spot tasks passed the accepted envelope through the final 100-jobs/60-minute window.

Evidence: https://github.com/Ontos-AI/knowhere-api-infra/issues/22#issuecomment-5302077433

Verification

• uv run pytest deploy/ecs/test_render_task_definitions.py -q — 7 passed
• uv run --group lint ruff check deploy/ecs/test_render_task_definitions.py deploy/ecs/render_task_definitions.py — passed
• git diff --check — passed

A repository-wide uv run pytest -q was also attempted, but collection requires optional workspace dependencies and runtime configuration absent from the isolated worktree (shared, fakeredis, fitz, syntok, and S3/database settings). The complete ECS deployment contract suite passes.

Deployment

This PR changes only the deployment source and its contract. It does not update an ECS service or create/remove any AWS resource.