Use shim for legacy client mutations batcher

[daniel-sanche]

This will likely be the most difficult functionality to shim, because of many small differences in the ways background threads are handled. Because we're making a breaking change though, we don't have to have all functionality 100% aligned, with the old version, and should aim for simplicity and stability.

This update will probably require many notes in the UPGRADING.md Migration Guide

Will require some work in the data client to support manual flush

See go/bigtable_legacy_deprecation for more details