HadoopInputFormatIO should support custom key/value coder

Right now HadoopInputFormatIO infers coders with getDefaultCoder but this doesn't work for cases like AvroCoder with GenericRecord, for example, when Avro is used with Parquet.

https://github.com/apache/beam/blob/master/sdks/java/io/hadoop/input-format/src/main/java/org/apache/beam/sdk/io/hadoop/inputformat/HadoopInputFormatIO.java#L288