Our document parsing code treats:
```
{"foo.bar": "baz"}
```
the same as:
```
{"foo": {"bar": "baz"}}
```
but the code that extracted routing_path didn't! This is bad because
routing_path has to identify a subset of dimensions precisely and in the
same way that our mapping code identifies the dimensions. They have to
line up or two time series could end up on different shards. Sad times.

This makes `routing_path` interpret dots in field names as though they
were an object. It creates an option for the includes/excludes filters
on `XContentParser` to treats dots as objects. So the filter would see
```
{"foo.bar": "baz"}
```
as though it were:
```
{"foo": {"bar": "baz"}}
```

So the filter `foo.bar` would match both documents.

This is part of elastic#82511.