Point workflow yaml at new docker image

[gburgessiv]

WIP; docker image hasn't landed yet (see #133137), but I want to see what wins can be verified/bugs can be found outside-of-production.

[gburgessiv]

I want to see what wins can be verified/bugs can be found outside-of-production.

• Initial pull/unpack of a docker image is regressed by about 60s on our builders. This is done once per builder at startup, so is likely neither here nor there.
• With a new LLVM checkout baked into the image, the 'sync' step sits around 15s.
• With a 5wk old LLVM checkout baked into the image, the 'sync' step sits around 32s.
• Without any LLVM checkout baked into the image, the 'sync' step sits around 75s.

So this saves somewhere from 40-60 seconds per run on Linux, assuming we expect to build new images roughly monthly.

I'll test Win next, but shaving the better part of a minute of per-run init off in exchange for 60s of VM startup time sounds like a win to me. Sounds like we might be able to make this docker image part of our cloud image at some point, which should take the VM startup cost to near-0.