rfc13: clarify KVS key constraints

[garlick]

While running down a problem with recent mpich mpiexec.hydra not being able to launch flux (flux-framework/flux-core#6072) a discussion with mpich developers over in pmodels/mpich#7050 resulted in some new knowledge about the PMI-1 wire protocol that we should document and change in our implementation.

Specifically:

• PMI KVS values cannot contain SPACE
• PMI_ prefixed PMI keys are a reserved namespace
• when a key in the reserved namespace cannot be fetched, the server returns success with an empty string

This PR adds some clarification on those topics to RFC 13.

[garlick]

Well when I added these restrictions to our PMI client, I found that openmpi doesn't work in CI. Sigh.

Apparently they try to put a string with a space in it. So this perhaps cannot be enforced at the API level. More like we'll just have to encode everything to ensure it doesn't break the wire protocol. Ironic since openmpi is already doing that. At least their content doesn't seem human readable.

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 -

[garlick]

Yeah I'm just going to close this and leave things as they are. :-(