Fail fast on creating mutations that cannot be stored #1919
Description
This is in the context of Beam CloudBigtableIO. The client has a bulk mutation limit of 100k. As of #1497 this means as long as each individual Put is under that limit everything will be fine.
A Dataflow job can have large variability in the amount of data per row. It is also natural to write processElement using a single Put, especially since that is what the examples guide new users to do.
If a single Put is larger than 100k mutations then it is impossible to store. The exception it causes is thrown later, and from a different thread.
The client should throw the TooManyMutations exception much earlier, preferably so the user code that caused the 100001th mutation it is still in the stack trace.