AverageSizeHint is too small for long strings

void IDataType::updateAvgValueSizeHint(const IColumn & column, double & avg_value_size_hint)
{
    /// Update the average value size hint if amount of read rows isn't too small
    size_t column_size = column.size();
    if (column_size > 10)
    {
        double current_avg_value_size = static_cast<double>(column.byteSize()) / column_size;

        /// Heuristic is chosen so that avg_value_size_hint increases rapidly but decreases slowly.
        if (current_avg_value_size > avg_value_size_hint)
            avg_value_size_hint = std::min(1024., current_avg_value_size); /// avoid overestimation
        else if (current_avg_value_size * 2 < avg_value_size_hint)
            avg_value_size_hint = (current_avg_value_size + avg_value_size_hint * 3) / 4;
    }
}

So we have the above function to update the size hint. CH mainly uses it for reader. However, when it is used in writer (DMFileWriter), this actually discards the useful info.

Start with IDataType::updateAvgValueSizeHint and trace how its size hint is used by readers and by DMFileWriter. Compare the writer behavior for long strings with the intended preservation of useful sizing information, then verify the corrected behavior with appropriate coverage.