Bigtable: race condition in resuming ReadRows can lead to unintentional full table scan

This client seems to have a similar bug as:

https://github.com/googleapis/google-cloud-go/commit/566ccf326f970b4fd1168bc0b6b474e94eeb6460
and
https://github.com/googleapis/python-bigtable/pull/554

I believe the resumption code:

https://github.com/googleapis/google-cloud-php/blob/6ece84084e1211afba977b3dbefd0b82730dd600/Bigtable/src/ChunkFormatter.php#L211-L222

Is vulnerable to a race condition:
1. client sends a read request for range [a,c]
2. server sends back the data chunks row c
3. but before the OK trailers are received, the deadline expires
4. the retry logic will craft a retry request that will exclude ranges that can been consumed, leaving an empty RowSet
5. client will use the empty RowSet for the retry attempt
6. server will interpret the RPC as a full table scan