[BUG] String and Char coordinates incorrectly written as variables with wrong dtype

[rhaegar325]

Describe the bug

When writing datasets with string/character coordinates (e.g., type coordinate in land variables), the current write() function has two issues:

1. String coordinates become variables: Coordinates like type are written as data variables instead of remaining as coordinates
2. Incorrect dtype: The dtype changes from |S11 (byte string) to <U11 (Unicode string), which is not CMIP6 compliant

Expected Behavior

String coordinates should:

• Remain as coordinates (appear in ds.coords, not ds.data_vars)
• Use CF-compliant character encoding with dtype='|S1' and a strlen dimension
• Match the encoding of standard CMIP6 datasets

Expected encoding:

{
    'dtype': dtype('S1'),
    'char_dim_name': 'type_strlen',
    'original_shape': (11,)
}

Actual Behavior (Before Fix)

Before writing:

ds.coords:
    type     |S11 11B b'bare_ground'  # Coordinate

After writing and re-reading:

ds.data_vars:
    type     <U11 'bare_ground'  # ❌ Now a variable, wrong dtype

ds.coords:
    # ❌ type is missing here

Root Cause

The write() function uses netCDF4 library directly but doesn't properly handle string coordinates:

1. Doesn't distinguish between string coordinates and regular variables
2. Doesn't apply CF-compliant character array encoding (S1 + strlen dimension)
3. Doesn't add string coordinates to the main variable's coordinates attribute (required for auxiliary/scalar coordinates per CF conventions)

[rhaegar325]

I will update the write() function to address this. Once this is resolved, issue #166 should be resolved as well.