It can be solved, for example, by specifying the schema with ID as string.

If you don't want to specify schema, better would be autoloader as you can specify hint:

spark.readStream
  .format("cloudFiles")
  .option("cloudFiles.format", "csv")
  .option("cloudFiles.schemaLocation", checkpoint_path)
  .option("cloudFiles.schemaHints", "ID String")
  .option("inferSchema", True)
  .option("mergeSchema", True)
  .load(folder)