Do Not Try To Install Benchmark Dependencies If Already Done

Currently we install each benchmark's requirements every time we run the benchmarks, which adds to the time it takes. If we know they are already installed then we can skip that.

Possible solutions:

• only install requirements when a venv is created
  • for the common venv, we'd need to track (in a file) which benchmarks have been done already
• only install requirements if not done already for the venv
  • track with a file (e.g. <venvroot>/.installed)
  • use a file to track what has already been installed in the venv
  • use pip freeze (or similar) to identify what has been installed already (requires the ability to verify requirements)

All that said, I'm not sure this alone is worth the effort. It might be worth it if combined with #157.

Start by tracing where benchmark requirements are installed and how the common and per-benchmark virtual environments are created. Compare tracking installed requirements with checking the environment, including the interaction with issue #157. Done means repeated benchmark runs skip work when dependencies are already available without compromising dependency correctness.