Connector runs out of memory if snapshot is too large

When the `use_snapshots` config property is `true`, the connector buffers an entire snapshot in memory before propagating to Kafka. This can be a problem if the first snapshot is huge, as it often is when streaming from the beginning of history.

The temporary workaround is to set `use_snapshots` to `false`.

A possible fix would be to treat the first snapshot as a special case, and never buffer it.