cluster: add schedulingPolicy to settings

[yashLadha]

Added schedulingPolicy option to settings so that the cluster owners
can schedule the handler explicitly as well. They can use the name
either rr or none for setting up the schedulingPolicy of
the primary process.

[mscdex]

Should we be keeping the assert() to ensure settings.schedulingPolicy maps to a supported value in SCHEDULE_POLICY_NAME_MAP?

[Mifrill]

As I understand, it should be handled via SafeMap above:
https://github.com/yashLadha/node/blob/9027b55935784c5e40ac54db5919895ad1636f16/lib/internal/cluster/primary.js#L47-L49
so the value of settings.schedulingPolicy can be only supported in SCHEDULE_POLICY_NAME_MAP. Hence the assert is redundant in this particular case.

[mscdex]

We need to keep this (cluster.schedulingPolicy) and consider its value to avoid breaking things.

[aduh95]

That's adding an option, so it's semver-minor, right? We'd need to add an entry in the changes list in the YAML comment if that's the case.

[mscdex]

I think it makes more sense to have settings.schedulingPolicy use the direct values like cluster.schedulingPolicy. The reason we have the string representations is to support setting from process.env.

Additionally, I think the changes here can be simplified by leaving the original top-level initialization of (cluster.)schedulingPolicy and then simply having schedulingPolicy: options.schedulingPolicy || cluster.schedulingPolicy in settings. Then we would just do schedulingPolicy = settings.schedulingPolicy and leave the pre-existing assert() to ensure schedulingPolicy holds a valid/supported value.