A suggested solution to the `TypeError: Invalid value for attr:` error upon `.to_netcdf`

Even though netcdf conventions don't allow some data types in the attributes, it might be usefull to simply serialize those as strings rather than throw an error. Maybe add a force_serialization keyword argument to the .to_netcdf method.

Example:
Setup a DataArray with bad values in the attributes:

import numpy as np
from xarray import DataArray, load_dataset
from pandas import Timestamp

from numbers import Number

valid_types = (str, Number, np.ndarray, np.number, list, tuple)

da = DataArray(
    name='bad_values',
    attrs=dict(
        bool_value=True,
        none_value=None,
        datetime_value=Timestamp.now()
    )
)

ds = da.to_dataset()
ds.bad_values.attrs

Output:

{'bool_value': True,
 'none_value': None,
 'datetime_value': Timestamp('2020-02-03 10:53:02.350105')}

The code in the except clause can be easily impolemented under _validate_attrs.

try:
    ds.to_netcdf('test.nc')
    # Fails with TypeError: Invalid value for attr: ...
except TypeError as e:
    print(e.__class__.__name__, e)
    for variable in ds.variables.values():
        for k, v in variable.attrs.items():
            if not isinstance(v, valid_types) or isinstance(v, bool):
                variable.attrs[k] = str(v)

    ds.to_netcdf('test.nc')  # Works as expected

ds_from_file = load_dataset('test.nc')
ds_from_file.bad_values.attrs

Output:

TypeError Invalid value for attr: None must be a number, a string, an ndarray or a list/tuple of numbers/strings for serialization to netCDF files

{'bool_value': 'True',
 'none_value': 'None',
 'datetime_value': '2020-02-03 10:43:38.479866'}

Start by running the provided DataArray and to_netcdf example to reproduce the invalid-attribute error. Then read _validate_attrs in xarray/backends/api.py and determine the intended force_serialization behavior, including the round-trip result shown in the issue; the issue does not name a test file.