[WIP] Remove our special duckdb installing

[jonkeane]

We've updated the way that duckdb gets used so that it is only needed to (re)create tpc-h data, and thus should be able to be installed only when that happens.

[jonkeane]

Created a test run: https://buildkite.com/apache-arrow/arrow-bci-benchmark-build-test/builds/72

[ElenaHenderson]

Thank you for this change and scheduling a test build!

[jonkeane]

Ok, looking at that build, there are two failures:

• (1) tpc-h seems to be failing on validation (though this might be due to rounding | floating point precision?). I tried this out locally and could not reproduce (and the code I've altered in arrowbench has been running successfully for a while now on the actual benchmarks) So I think this is fine
• (2) it looks like generating scale factor 10 data also failed — this isn't totally surprising if the machine running this test doesn't have a huge amount of memory. Maybe we should disable that test for this test build?
• (3) micro c++ benchmarks failed — I'm not totally sure why, nothing I changed should impact them.

I suspect these issues are all from the test harness and not the change to the code — but would love a second opinion on that @ElenaHenderson

[ElenaHenderson]

@jonkeane I am pretty confident that builds failures are not caused by the changes in this PR.

I created #74 to fix environment used for running arrow-bci-benchmark-build-test builds.