Fix Hyper scripts to perform hot runs

[caetanosauer]

A recent commit aimed at "Prevent cheating by Munich-based databases" (4a2f3fa) incorrectly modified Hyper's query.py scripts to restart the database server on each iteration, essentially making every run a cold run. This unfairly brought our results down by a large margin.

This commit reverts the previous behavior which was already correct and there was absolutely no cheating involved.

[rschu1ze]

@caetanosauer There was some discussion about what "cold" runs really mean in #667. I pushed #692 in response. Tl;dr is that today's Hyper submission is fine - now we'll only need to migrate all other submissions one-by-one from lukewarm to cold runs for fair comparison. Allow me to close this PR - thanks.

PS: I removed "cheating" from the message of #659. This term was unnecessarily dramatic.

[caetanosauer]

@rschu1ze Thanks for your response and the update on the lukewarm definitions.

The problem I mentioned in this PR still persists, though: There are no hot runs being measured for Hyper. Last time I measured, Hyper was second place behind Umbra in the "Hot Run" category. Now we are way down below, because all 3 repetitions are run on a freshly restarted Hyper.

[rschu1ze]

@caetanosauer I see, let's re-open the PR.

However, I like to clarify with you if the first run is cold or lukewarm (see the PR / issue I linked above).

In

with HyperProcess(telemetry=Telemetry.DO_NOT_SEND_USAGE_DATA_TO_TABLEAU) as hyper:
    with Connection(hyper.endpoint, 'hits.hyper', CreateMode.NONE) as connection:
        for _ in range(3):
            start = timeit.default_timer()
            try:
                connection.execute_list_query(query)

it is not clear to me if the Hyper process survives between queries (or to be more precise: survives after three runs of a query).

We should physically terminate the database to ensure true cold runs.

[caetanosauer]

The with HyperProcess guard starts a new process and terminates it at the end. Since the script query.py is invoked for each query from run.sh after clearing the filesystem cache, the first iteration of each query will be truly cold.