Defaulting `max_local_storage_nodes` to `50` is useful for testing so
this patch moves that default from "all the time" to "if you don't bind
a public ip". If you *do* bind a public IP and attempt to use more than
a single node per data directory without *explicitly* setting
`max_local_storage_nodes` then Elasticsearch will now fail to start.

Since you can't change the default for a setting depending on the
"prod-mode/non-prod-mode" flag we instead use a sentinel value (`0`) of
`max_local_storage_nodes` to mean "user didn't specify, pick it from
the "prod-mode/non-prod-mode" flag".

Since we don't know if we're in prod mode when we pick the data directory
we instead assume that we *aren't* and then check if the directory that
we picked was OK when we run the `BootstrapCheck`s. The nice thing about
doing it this way is that we warn the user if they have more than one
Elasticsearch in their data directory even if they are running even in dev
mode!

While this is the easiest way to make this change the cost is that, even
if Elasticsearch fails to start because it is in production mode and
`max_local_storage_nodes` isn't configured it'll still create a directory.

This also removes some of the documentation that suggests you override the
setting on all production clusters. That is no longer important because
Elasticsearch won't start in production mode if you try to start more
than one node.

Closes elastic#19679