In some cases, we have the estimated row count hint for how many rows will be needed from the table reader. In the initial version of the KV projection pushdown work that information is not utilized, but it'd be nice to take advantage of the hint. In particular, this will allow the wrapped cFetcher to create batches of reasonable capacity (rather than doing exponential increase starting with the size of 1).

Note that this estimate is already used to set MaxSpanRequestKeys limit on the whole BatchRequest on the SQL side.

Jira issue: CRDB-23186