Reconsider `required` property for `process.*` semantic conventions

[arminru]

Currently we define process.pid and "at least one of process.executable.name, process.executable.path, process.command, or process.command_line" as required, if any resource attribute in the process namespace is present. I don't think this is a sensible requirement, since there are multiple concerns (or multiple features from a backend's perspective) mixed together in the process namespace. I could well imagine that one might want to report the runtime (added in #882) but has no need for reporting the PID and/or the entire commmand line. This would currently not be "legal", however.
I think we should drop the entire required column for the process namespace.

This is related to #653. I do, however, think of the trace semantic conventions more as distinct features, where defining some set of attributes as essential to provide some reasonable output makes more sense than for the process resource attributes.

[Oberon00]

I think we should just put process.runtime into it's own resouce type (without changing the names)

[Oberon00]

@open-telemetry/technical-committee I think this issue should be allowed-for-ga, maybe even P2. EDIT: I made a PR #1137 that would fix this.

[andrewhsu]

from the spec sig mtg today, re-triaged for allowed-for-ga