Fix rate limits under multi-node environments

What does this MR do?

Replace read-modify-write with atomic operations and optimistic locking to ensure that rate limits are correct when multiple threads / processes are in use. This changes the algorithm and performance characteristics, which are now documented in SkRateLimiterService.md.

Contribution Guidelines

By submitting this merge request, you agree to follow our Contribution Guidelines

• I agree to follow this project's Contribution Guidelines
• I have made sure to test this merge request