Allow adding custom types to _native_to_marshmallow

Follow up to #15, this is to suggest exposting _native_to_marshmallow publicly or, better yet, allow registring custom types to the list for more flexibility on the types that could be user-defined and which fields these types should translate to.

[alextremblay]

There was an idea introduced in another issue (#23) That I think could work very well here:

from typing import List
from ipaddress import IPv4Address

from marshmallow.fields import Field
from marshmallow_dataclass import dataclass, NewType

class IPField(Field):
    ... # serializes an IPv4Address instance into a str, and deserializes back into an instance

#IP = NewType('IP', str, validate=my_validation_function) 
#This was the suggestion in the other issue

IP = NewType('IP', IPv4Address, marshmallow_field=IPField)

@dataclass
class Company:
    name: str
    ipv4s: List[IP]

The idea is to have a NewType implementation that works the same as typing.NewType, but also stores all the field metadata that we would normally store on a dataclass field, like validation functions and marshmallow_field selections.

Thoughts?

[evanfwelch]

I would also find this valuable… a mapping you can can pass to class_schema that gets merged with _native_to_marshmallow.

    if typ in _marshmallow_mapping:
        field_fn = _marshmallow_mapping[typ]
        if inpsect.isfunction(field_fn):
            return field_fn(typ, **metadata)
        else: #assume it's just a marshmallow field
             return field_fn(**metadata)