Update skipped list for upstream CI

[i-chaochen]

Motivation

after this PR is merged openxla#36893 we are fully using rocm-dev-infra for upstream CI and we can better control the skipped list.

[i-chaochen]

let's wait this PR is merged first openxla#40702

[i-chaochen]

@amd-jianli12 seems this one is failed on your PR openxla#40702 as well?

//xla/backends/gpu/runtime:host_execute_thunk_test_amdgpu_any            FAILED in 3 out of 3 in 22.5s

does @Eetusjo work on this?

[Eetusjo]

I think there is some more general issue that makes certain tests flaky, the error is one we have seen with other tests as well: RESOURCE_EXHAUSTED: Failed to allocate managed memory: HIP_ERROR_OutOfMemory. E.g. we have this issue https://github.com/orgs/ROCm/projects/14/views/1?pane=issue&itemId=162453096&issue=ROCm%7Cframeworks-internal%7C15776 where //xla/stream_executor/rocm:rocm_executor_test is flaky with that error.

I just checked some CI runs from recent commits and didn't see host_execute_thunk_test failing so probably same flaky behavior.