Prevent lukewarm cold runs in databases from Munich

[alexey-milovidov]

Umbra, CedarDB, and Hyper have unrealistic results in cold runs due to excessive caching.

According to the rules:

It is okay if the system performs caching for source data (buffer pools and similar). If the cache or buffer pools can be flushed, they should be flushed before the first run of every query.

If the system contains a cache for intermediate data, that cache should be disabled if it is located near the end of the query execution pipeline, thus similar to a query result cache.

To ensure that the caches are flushed and data is actually read from disk during the cold run, I added the restart of the Docker container before the cold run of each query.

Also found that hyper-parquet didn't do a cache flush at all.

Note: fixing this problem was suggested by puzpuzpuz #656