Switch to larger Helios instances for CI

[emilyalbini]

This PR switches the slowest Helios jobs to run on larger instance sizes, significantly speeding up CI times. Along with this I already deployed a Buildomat configuration change to run all Helios jobs on Zen 4 AWS instances, instead of Zen 3 instances either on AWS or lab Gimlets. Together, these two changes should bring CI times down considerably.

Unfortunately we cannot use Zen 5 AWS instances (like we did on Linux) until oxidecomputer/stlouis#938 is fixed.

build-and-test (helios)

	Architecture	CPU cores	RAM	Execution time	Price per build
Old:	Zen 3	8	32 GB	135 minutes	$1.04
	Zen 4	8	32 GB	110 minutes	$0.85
	Zen 4	16	128 GB	58 minutes	$1.24
New:	Zen 4	32	256 GB	45 minutes	$1.87

This job was actually slowed down for a nondeterministic amount of time by it running out of memory and being forced to aggressively page memory to disk. Turns out it was using around 150% of the RAM the VM had allocated. Switching to memory-optimized AWS instances (2x the RAM) fixed the problem.

The switch from 16 cores to 32 cores is fairly expensive and has diminishing returns, like for the Linux instance, but still, it's a 15 minutes win. When we switch to Zen 5 it might be worth it to go back to 16 cores.

omicron-common

	Architecture	CPU cores	RAM	Execution time	Price per build
Old:	Zen 3	8	32 GB	6 minutes	$0.05
New:	Zen 4	8	32 GB	5 minutes	$0.04

The switch has negligible impact on a job this short, but it's not worth it to create a dedicated target just to keep this job back on Zen 3. So it gets unintentionally updated to Zen 4.

helios / package

	Architecture	CPU cores	RAM	Execution time	Price per build
Old:	Zen 3	8	32 GB	46 minutes	$0.35
	Zen 4	8	32 GB	30 minutes	$0.23
New:	Zen 4	16	64 GB	24 minutes	$0.37

The wins from 8 cores to 16 cores are not that impressive, but this job is a dependency of the "deploy" job which we cannot really speed up (it needs to run on a lab Gimlet, and we can't shard it as far as I'm aware), so any time we can shave is worth it.

helios / build TUF repo

	Architecture	CPU cores	RAM	Execution time	Price per build
Old:	Zen 3	8	32 GB	80 minutes	$0.62
New:	Zen 4	16	128 GB	43 minutes	$0.92

Similarly to the build and test job this was paging memory to disk due to not having enough memory in the VM (even though to a less extent). After the size increase there was a lot of single-thread CPU, so I didn't bother testing more cores.

check-features (helios)

	Architecture	CPU cores	RAM	Execution time	Price per build
Old:	Zen 3	8	32 GB	38 minutes	$0.29
	Zen 4	1	4 GB	55 minutes	$0.04
	Zen 4	2	8 GB	38 minutes	$0.07
New:	Zen 4	8	32 GB	24 minutes	$0.19

This job was mostly single-threaded so I tried aggressively reducing the VM size but with mixed results. In the end decided to keep it with the now-Zen4 standard target.

clippy (helios)

	Architecture	CPU cores	RAM	Execution time	Price per build
Old:	Zen 3	8	32 GB	26 minutes	$0.20
New:	Zen 4	8	32 GB	18 minutes	$0.14
	Zen 4	16	64 GB	17 minutes	$0.26

Turns out there was zero benefit going from 8 to 16 cores for this job.

[davepacheco]

Looks great (once you rebase onto #10001).