Support running multi-device tests in CI

[danielvegamyhre]

For float8 training, the test_everything.sh script requires multiple GPUs for FSDP/TP tests, so we currently don't run in CI as it's not configured for multi-device jobs. We should figure out how to run these multi-device tests in CI. This would also be useful for some of our new MoE training parallelism tests.

[jainapurva]

@danielvegamyhre We're hoping to get multi-gpu support for CI soon.

[danielvegamyhre]

@danielvegamyhre We've hoping to get multi-gpu support for CI soon.

I talked to Wouter and he said we could use linux.aws.h100.4 or linux.aws.h100.8 for multi-gpu tests and it seems to work so far