Enable CPU based throttling in Bigtable client

[dbolduc]

Sometimes batch write jobs want to optimize for CPU utilization, instead of minimizing latency or maximizing throughput.

At the moment, the servers can only communicate errors to us, such as kDeadlineExceeded. There is a way to set some feature flags and have the server return throttling information. They essentially tell us whether we should speed up the current QPS, or whether we should slow down the current QPS.

Phase 1

• Add RateLimiter
• Add bigtable-specific wrapper around the RateLimiter. I am calling it the MutateRowsRateLimiter. This thing has an UpdateRate(MutateRowsResponse const& r).
• Add a struct experimental::BulkApplyThrottlingOption { using Type = bool; };
• Plumb the MutateRowsRateLimiter through the client layers

Phase 2

• Tracing for sleeps
• AsyncBulkApply support