xarrayprovider gen_covjson fails with nan data

[dschneider-wxs]

Description
I am using the XarrayProvider and XarrayEDRProvider with a netcdf file and the XarrayProvider.gen_covjson() seems to be tripping over itself. If the dataset has missing values as NaN, they are converted to None and then fail during serialization because np.isnan on None raises a TypeError.

starting at Line 400 pygeoapi/provider/xarray_.py

        data = data.fillna(None)

        try:
            for key, value in selected_fields.items():
                LOGGER.debug(f'Adding range {key}')
                range = {
                    'type': 'NdArray',
                    'dataType': value['type'],
                    'axisNames': [
                        'y', 'x'
                    ],
                    'shape': [
                        metadata['height'], metadata['width']
                    ],
                    'values': [
                        None if np.isnan(v) else v
                        for v in data[key].values.flatten()
                    ]
                }

I don't think I understand the point of setting None here if the code is checking for nan.

[franko-c]

if I'm reading this right, xarray_.py:400 calls data.fillna(None) which converts NaN floats to None — then :415 runs np.isnan(v) on the flattened values and gets TypeError because None isn't numeric. removing the fillna should work since the comprehension at :415 already handles NaN→None.

[dschneider-wxs]

yep, i have a local branch that seems to work without L400.