createReadStream() creates a pipeline of streams that converts a stream of row chunks into a stream of logical rows. It also has logic to handle stream resumption when a single attempt fails.
The pipeline can be split into 2 parts: the persistent operation stream that the caller sees and the transient per attempt segment. When a retry attempt occurs, the per attempt segment is unpiped from the operation stream and is discarded. Currently this includes any buffered data that each stream might contain. Unfortunately, when constructing the retry request, createReadStream() will use the last row key from the last buffered row. This will cause the buffered rows to be omitted from the operation stream.

This PR fixes the missing rows part by only referencing the row keys that were seen by the persistent operation stream when constructing a retry attempt. In other words, this will ensure that we only update the lastSeenRow key once the row has been "committed" to the persistent portion of the pipeline

# Conflicts:
#	system-test/read-rows.ts

# Conflicts:
#	system-test/read-rows.ts

Co-authored-by: Leah E. Cole <[email protected]>

…ermark-removal' of https://github.com/danieljbruce/nodejs-bigtable into fix-missing-rows-with-test-and-fix-for-node-14-plus-watermark-removal

Use it to replace any