Improve cross-crossgen run time

[jashook]

Our cross-crossgen jobs take ~10 minutes to complete, and hold one linux arm machine for the entire time. The run time can be improved by parallelizing the crossgen invocations. This should crossgen in roughly ~4x the time due to the machines having four cores.

This will hopefully take our job time down to ~5 minutes.

[jashook]

Time in seconds

Branch	x64-arm crossgen (build machine)	arm crossgen (on device)
master	189.140369	448.044597
#33175	87.963722	133.222282

[jashook]

After #33175 goes in crossgen2 should hopefully be able to improve further. However, the low hanging fruit is no longer there.

[jashook]

To improve the machine utilization further I see a few different possibilities.

1. Before submitting to helix, remove all x64 binaries in the core_root (Crossgen comparison job, remove x64 binaries before submitting #34103)
2. Remove all build related dependencies from the helix docker images. (Remove all build related dependencies from the helix docker images #34104)