mzbench: remove

This is just a proposal for now, but I wanted to surface the amount of
complexity and duplication in mzbench. What pieces do we need to keep?
What I've heard is:

  * A replacement for the avro-insert/avro-upsert benchmarks that is
    easy to use locally.

  * Parallelization of kgen to replace the kafka-avro-generator wrapper.

Author: benesch

bin/mzbench

@@ -186,33 +186,6 @@ steps:
           run: ci
     timeout_in_minutes: 30
 
-  - id: aggregations-benchmark
-    label: "aggregations benchmark sanity check"
-    depends_on: build-x86_64
-    plugins:
-      - ./ci/plugins/mzcompose:
-          composition: aggregations
-          run: ci
-    timeout_in_minutes: 10
-
-  - id: avro-upsert-benchmark
-    label: "avro upsert benchmark sanity check"
-    depends_on: build-x86_64
-    plugins:
-      - ./ci/plugins/mzcompose:
-          composition: avro-upsert
-          run: ci
-    timeout_in_minutes: 10
-
-  - id: kafka-sink-avro-debezium-benchmark
-    label: "kafka sink (avro debezium) benchmark sanity check"
-    depends_on: build-x86_64
-    plugins:
-      - ./ci/plugins/mzcompose:
-          composition: kafka-sink-avro-debezium
-          run: ci
-    timeout_in_minutes: 10
-
   - id: tb
     label: tb
     depends_on: build-x86_64