Benchmark downloader script should take the latest score, not the smallest

[Akirathan]

Thanks to the observations of @JaroslavTulach, I have noticed that on our benchmark results pages (for both engine and stdlib), we have a lot of nonsensical data.

Problem description

Every benchmark run generates bench-results.xml file with a format like this:

        <case>
            <label>org.enso.benchmarks.generated.Startup.empty_startup</label>
            <scores>
                <score>2823.877618</score>
            </scores>
        </case>

If the runner is not clean, the bench-results.xml file is not rewritten, but rather a new score value is appended, like so:

        <case>
            <label>org.enso.benchmarks.generated.Startup.empty_startup</label>
            <scores>
                <score>2823.877618</score>
                <score>3206.157912</score>
            </scores>
        </case>

The procedure in the bench_download.py script used to choose the lowest of these numbers:
https://github.com/enso-org/enso/blob/develop/tools/performance/engine-benchmarks/bench_download.py#L280-L284
This is obviously wrong. This PR fixes this by choosing the latest score, that is the last value.

TL;DR;

All our benchmark data from runners that were not clean are wrong! On this picture, on the right there is the old and wrong value, and on the left is the correct one:

The data was fetched from this snippet:

[Akirathan]

I have just regenerated the results on the servers on https://enso-org.github.io/engine-benchmark-results/. They are now correct and up-to-date.

My pageEngine and Standard libs benchmarks

[JaroslavTulach]

Taking the last value is a good hotfix for now, but we need also fix in the CI. There is a removal code at

enso/build/build/src/engine/context.rs

Line 226 in 4983550

// Remove the benchmark reports. They are not meant currently to be incrementally

but it only works for engine/runtime benchmarks.

We need to make sure it also works for std-benchmarks. Why not remove bench*xml in the whole repository, @mwu-tow?