[FEA] Support basic schema evolution for AVRO reading

[sperlingxx]

Is your feature request related to a problem? Please describe.
Currently, we can't read more columns than the file schema of AVRO, which is supported under PARQUET/ORC reading.

def test_read_avro_with_empty_clipped_schema(spark_tmp_path, v1_enabled_list, reader_type):
    data_path = spark_tmp_path + '/AVRO_DATA'
    with_cpu_session(
        lambda spark: gen_df(spark, [('a', int_gen)], length=100
                             ).write.format("avro").save(data_path))
    schema = StructType([StructField('a', IntegerType()), StructField('c', StringType())])
    assert_gpu_and_cpu_are_equal_collect(
        lambda spark: spark.read.schema(schema).format("avro").load(data_path),
        conf=all_confs)

To support this feature, we need to add transformations to fill the gap between read schema and file schema, just like what we did for PARQUET reading and ORC reading.