set_index to date with dask dataframe to circumvent memory error

Question

I have a very large pandas dataset and when I try to use set_index, I get an out-of-memory error, which I can somehow understand. I would like to transform the datetime into just giving me the date for the index, which works normally with df.set_index(df.index.date, inplace=True).

I thought I could work around this using dask.dataframe, but I haven't found a way to get it working. I have tried different methods I found on SO (I will link them when I find them again) but to be honest, I do not understand well the what I'm doing there. Any help much appreciated!

    import numpy as np
    import dask.dataframe as dd
    import pandas as pd
    import datetime as dt
    
    def make_df(date0,x,y):
        ''' simulate my data (normally much larger) '''
        
        xx,yy = np.meshgrid(x,y)
        zz    = np.sin(xx/20) *np.cos(yy/30)
        tt    = pd.date_range(date0, freq='92min', periods=zz.size)

        arr      = np.full((xx.size,3),np.nan)
        arr[:,0] = xx.flatten()
        arr[:,1] = yy.flatten()
        arr[:,2] = zz.flatten() + np.random.randn(zz.size) / 10

        df = pd.DataFrame(arr, columns=['x','y','z'], index=tt)
        df.index.name = 'time'

        return df
    
    # create dataset
    dist  = 2
    date0 = dt.datetime(2023, 9, 23)
    x     = np.arange(-10,60,dist)
    y     = np.arange(30,70,dist)
    df    = make_df(date0,x,y)
        
    df = dd.from_pandas(df, npartitions=150)
    print(df)
    
    ### ValueError: Metadata inference failed in `lambda`.
    # df.map_partitions(lambda x:
    #                   x.set_index(df.index.dt.date)
    #                   ).persist()
    
    ### AttributeError: 'Index' object has no attribute 'dt'
    def func(x):
        return pd.to_datetime(x.index.dt.date, utc=True)
        
    meta = pd.Series([], name='time')
    df['time'] = df.apply(func, axis=1, meta=meta)
    
    df = df.set_index('time', #drop=False, compute=True
                      )
    
    df.compute()
    print(df)

    return