Error generating chart: Collection.map: A mapped algorithm must return a Feature or Image

I am trying to create a timeseries chart in GEE and I keep having an error:

Error generating chart: Collection.map: A mapped algorithm must return a Feature or Image

What can be the reason for this error and how can I solve it ?

I am trying to create a kNDVI timeseries from each cluster on the site.

    var site = 
    /* color: #0b4a8b */
    /* displayProperties: [
      {
        "type": "rectangle"
      }
    ] */
    ee.Geometry.Polygon(
        [[[14.189235213951225, -18.464648340471996],
          [14.189235213951225, -19.8242220793087],
          [18.770534042076225, -19.8242220793087],
          [18.770534042076225, -18.464648340471996]]], null, false);

// 2. WorldCover and grassland mask
var worldcover = ee.Image('ESA/WorldCover/v200/2021');
var grassMask = worldcover.eq(30).clip(site);

// 3. Cloud/shadow mask function
function maskS2clouds(image) {
  var scl = image.select('SCL');
  var mask = scl.neq(3)  // 3 = cloud shadow
    .and(scl.neq(7))    // 7 = unclassified
    .and(scl.neq(8))    // 8 = cloud medium prob
    .and(scl.neq(9))    // 9 = cloud high prob
    .and(scl.neq(10))   // 10 = thin cirrus
    .and(scl.neq(11));  // 11 = snow
  return image.updateMask(mask);
}

// 4. Band scaling and kNDVI calculation
function addKNDVI(image) {
  // Scale bands
  var bands = ['B2','B3','B4','B5','B6','B7','B8','B8A','B11','B12'];
  var scaled = image.select(bands).divide(10000);
  image = image.addBands(scaled, null, true);
  
  // Calculate kNDVI
  var RED = image.select('B4');
  var NIR = image.select('B8');
  var sigma = 0.15;
  var kndvi = NIR.subtract(RED).pow(2)
    .divide(sigma * sigma * 4.0)
    .tanh()
    .rename('kndvi');
  return image.addBands(kndvi);
}

// 5. Sentinel-2 ImageCollection
var s2 = ee.ImageCollection('COPERNICUS/S2_SR_HARMONIZED')
  .filterBounds(site)
  .filterDate('2019-01-01', '2019-12-31')
  .map(maskS2clouds)
  .map(addKNDVI);

// 6. Create composite and mask to grassland
var composite = s2.select(['kndvi']).mean().clip(site);
var compositeGrass = composite.mask(grassMask);

// 7. Perform clustering
var training = compositeGrass.sample({
  region: site,
  scale: 10,
  numPixels: 1000,
  seed: 1
});

var clusterer = ee.Clusterer.wekaKMeans(15).train(training);
var clustered = compositeGrass.cluster(clusterer);

// 8. Time Series Analysis - NOW PROPERLY DEFINED
var timeSeriesData = s2.map(function(image) {
  // Add cluster band to each image
  var withCluster = image.select('kndvi').addBands(clustered.select('cluster'));
  
  // Calculate mean kNDVI per cluster
  var stats = withCluster.reduceRegion({
    reducer: ee.Reducer.mean().group({
      groupField: 1,
      groupName: 'cluster'
    }),
    geometry: site,
    scale: 10,
    maxPixels: 1e9
  });
  
  // Format results
  var date = ee.Date(image.get('system:time_start')).format('YYYY-MM-dd');
  var properties = {
    'date': date,
    'system:time_start': image.get('system:time_start')
  };
  
  // Add cluster means
  var groups = ee.List(stats.get('groups'));
  for (var i = 0; i < 15; i++) {
    var clusterData = ee.Dictionary(groups.get(i));
    var clusterNum = clusterData.get('cluster');
    var meanVal = clusterData.get('mean');
    properties['cluster_' + clusterNum] = meanVal;
  }
  
  return ee.Feature(null, properties);
});

// 9. Create and display chart
var chartData = timeSeriesData.map(function(feature) {
  var date = feature.get('date');
  var props = feature.toDictionary();
  
  return ee.List.sequence(1, 15).map(function(clusterNum) {
    return ee.Feature(null, {
      'date': date,
      'cluster': 'Cluster ' + clusterNum,
      'kndvi': props.get('cluster_' + clusterNum)
    });
  });
}).flatten();

var chart = ui.Chart.feature.groups({
  features: ee.FeatureCollection(chartData),
  xProperty: 'date',
  yProperty: 'kndvi',
  seriesProperty: 'cluster'
}).setOptions({
  title: 'kNDVI Time Series by Cluster',
  hAxis: {title: 'Date'},
  vAxis: {title: 'kNDVI'},
  lineWidth: 1,
  pointSize: 3
});

print(chart);

// 10. Export data
Export.table.toDrive({
  collection: timeSeriesData,
  description: 'Cluster_kNDVI_TimeSeries',
  fileFormat: 'CSV',
  selectors: ['date', 'system:time_start'].concat(
    ee.List.sequence(1, 15).map(function(i) {return 'cluster_' + i;})
  )
});

// 11. Display layers
Map.centerObject(site, 8);
Map.addLayer(clustered.randomVisualizer(), {}, 'Clusters');
Map.addLayer(site, {color: 'black'}, 'Study Area');
Map.addLayer(grassMask.updateMask(grassMask), {palette: ['green']}, 'Grassland Mask');