Reformatting Dictionary Objects to rows and columns in Google Earth Engine

There's two rules of efficient Earth Engine programming that will not only make your script run better on large data sets, but will also make it simpler to get the result you want.

1. Instead of using reduceRegion several times on the same region with different reducers, construct one combined reducer that computes all the results together (in one output dictionary).
2. Process collections as collections; avoid toList and toBands. This way, the results for each month are separate features in the output collection, not entries in a dictionary. Since feature collections are tables, this means you don't have to do any data conversion to export them.

var dataset = ee.ImageCollection('IDAHO_EPSCOR/TERRACLIMATE')
  .filterBounds(subwatershed)
  .filterDate("1979-01-01","2016-01-01")
  .select("tmmx");

var results = dataset.map(function (image) {
  var scaledImage = image.divide(10);
  var reduction = scaledImage.reduceRegion({
    reducer: ee.Reducer.minMax().combine({
      reducer2: ee.Reducer.mean(),
      sharedInputs: true,
    }),
    geometry: subwatershed.geometry(),
    scale: 30,
  });
  // Attach the reduction results and month as properties to the features.
  return image.setMulti(reduction)
    .set('month', image.date().get('month'));
});

// Preview
print(results.limit(10));

// Export CSV
Export.table.toDrive({
  collection: results,
  selectors: ['month', 'tmmx_mean', 'tmmx_min', 'tmmx_max'],
  fileFormat: 'csv',
  description: 'tmmx_statistics',
});

https://code.earthengine.google.com/1f7dec81dcbd6ebb52157cf0ee423779

This will produce table output like the following. (Note that your ROI asset wasn't public, so I drew a polygon just for testing and the numbers won't match the ones you want.)