Acked-by: Tapani Pälli <tapani.palli@intel.com>

Acked-by: Tapani Pälli <tapani.palli@intel.com>

This changes all benchmark sources to a local cache on the Mesa perf CI
master system. Grabbing them remotely or from a system on a different
Mesa CI network causes failures if network utilization is high.

Untarring the benchmark will hang on jenkins if this is not passed.

gfxbench5 also supports manhattan, trex, fill, alu2, driver2 and egypt
benchmarks.

ModuleNotFoundError is first introduced in python 3.6 which is a bit
newer than we want to require.

The Vulkan bits unfortunately don't work with --mesa-path right now.
That needs to be sorted out but it will probably require refactoring of
a decent chunk of sixonix.

It's simple but fairly effective.  Currently, it only completes for the
subcommand and the benchmark name when using "run".

At this point, the python framework has a large chunk of what the old
shell scripts had.  There is still some functionality missing but it can
be dug out of the git history and ported over to python on an as-needed
basis.

The only subcommand we have for now is "run".  This replaces
run_benchmark.py

This version is slightly different than the original in that it writes
out a CSV file instead of dumping a pile of files into the current
directory.

We used to print it from run, now that's the caller's job.

It's a bit nasty to add a function to a module after loading it but it
saves us a lot of typing.