Update known Cray MPICH issues

[msimberg]

• Add less drastic workaround for hangs (with usually better performance, but doesn't work for all applications)
• Add note about slow intra-node communication and workaround to use NICs for intra-node communication

I'm going through our old lists of known issues. I have to check whether the other workarounds/issues are still relevant, so starting with what we know is still relevant.

[github-actions]

preview available: https://docs.tds.cscs.ch/256

[jpcoles-cscs]

The cxil_map error now appears three times, and it also marked as resolved. I think we can remove the first two occurrences? Sometimes I've had this error when using the wrong combination of libfabric and the aws ofi plugin.
The slow comm part is fine.

[github-actions]

preview available: https://docs.tds.cscs.ch/256

[msimberg]

The cxil_map error now appears three times, and it also marked as resolved. I think we can remove the first two occurrences? Sometimes I've had this error when using the wrong combination of libfabric and the aws ofi plugin. The slow comm part is fine.

Good catch! The one that I just added I've removed. That was a copy-paste mistake and wasn't meant to be duplicated at all...

The resolved and the unresolved cxil_map errors are actually not the same: the first was the system misconfiguration, and the second is one that has been spotted "occasionally" even after the system misconfiguration was fixed. But perhaps either the titles could be changed, or the resolved one can be removed altogether now? It's been fixed for a year now.