…sh`.

Before, the benchmark used 1 partition.  Now it uses 16 by default, and
`--partitions` can change that.

The current difference in performance in my testing, for 1 versus 16
partitions with 10M events with 512 kB messages (which is what our
generator uses), is within the margin of error.

Signed-off-by: Ben Pfaff <blp@feldera.com>

Signed-off-by: Ben Pfaff <blp@feldera.com>

…ges.

Kafka input performance for small messages was previously much worse than
`rpk`, the Redpanda utility for working with Kafka topics.  This commit
introduces the ability for the input adapter to use multiple threads to
read an input topic.  This substantially improves the performance for
small messages, bringing it much closer to `rpk` performance.

The following table shows the performance with default settings before and
after this commit with various message sizes and with `rpk`:

```
╭──────────────────────────────┬──────────────────╮
│                              │      method      │
│                              ├──────┬─────┬─────┤
│                              │before│after│ rpk │
├──────────────────────────────┼──────┼─────┼─────┤
│partitions 1  size 256    time│ 40.81│13.89│10.68│
│                   512    time│  8.63│ 6.26│ 5.34│
│                   4096   time│  1.68│ 1.76│ 2.15│
│                   65536  time│  2.01│ 1.99│ 1.96│
│                   524288 time│  1.62│ 1.64│ 2.08│
│          ╶───────────────────┼──────┼─────┼─────┤
│           16 size 256    time│ 51.97│14.32│     │
│                   512    time│ 13.69│ 7.08│     │
│                   4096   time│  1.81│ 2.12│     │
│                   65536  time│  1.65│ 1.74│     │
│                   524288 time│  1.64│ 1.87│     │
╰──────────────────────────────┴──────┴─────┴─────╯
```

Signed-off-by: Ben Pfaff <blp@feldera.com>