IO: Improve BulkProcessor when running per-record operations

[amotl]

About

When batches are of size one, SQLAlchemy relays a single insert operation to CrateDB, not a bulk one. This patch handles that situation better, effectively checking rowcount for handling INSERT OK, 0 rows responses.

Background

products.json#L11 demonstrates a record that can't be inserted into CrateDB. The behavior is that CrateDB doesn't fail on it, but responds with rowcount: 0.

crash -c "drop table if exists t01;"
crash -c "create table t01 (data OBJECT(DYNAMIC));"

http "127.0.0.1:4200/_sql?error_trace=false" \
  stmt="INSERT into t01 (data) values (?)" \
  args:='[{"_id": "507d95d5719dbef170f15c22", "item": [{"amount": { "percent_of_service" : 0.06 }}, { "amount" : 2.25 }]}]' --print bB

crash -c "refresh table t01;"
crash -c "select * from t01;"