feature benchmark: add scenario with many Kafka sources in the same cluster

[nrainer-materialize]

Nightly: https://buildkite.com/materialize/nightly/builds?branch=nrainer-materialize%3Afeature-benchmark%2Fmany-kafka-sources

[nrainer-materialize]

I am not 100% sure if this tests what we intend. I am a bit surprised by the low wallclock duration, which is < 10 seconds...
It is either extremely fast or some relevant logic happens already outside of /* A */ [...] /* B */.

[def-]

Did you try increasing the scale? If with a larger scale it gets slower I'm ok with this.

[nrainer-materialize]

There is a difference when using more sources.

With scale 2.5 (316 sources):

ManyKafkaSourcesOnSameCluster       | wallclock       |          25.569 |          24.310 |      no       | 5.2 pct   more/slower
ManyKafkaSourcesOnSameCluster       | messages        |            None |            None |      no       | N/A
ManyKafkaSourcesOnSameCluster       | memory_mz       |        4684.448 |        4555.702 |      no       | 2.8 pct   more/slower
ManyKafkaSourcesOnSameCluster       | memory_clusterd |          64.383 |          61.665 |      no       | 4.4 pct   more/slower

With scale 2.0 (100 sources):

ManyKafkaSourcesOnSameCluster       | wallclock       |           9.057 |           8.760 |      no       | 3.4 pct   more/slower
ManyKafkaSourcesOnSameCluster       | messages        |            None |            None |      no       | N/A
ManyKafkaSourcesOnSameCluster       | memory_mz       |        2271.652 |        2024.651 |    !!YES!!    | 12.2 pct   more/slower
ManyKafkaSourcesOnSameCluster       | memory_clusterd |          63.248 |          63.381 |      no       | 0.2 pct   less/faster