streaming方式的CombineFileInputFormat实现

hadoop版本问题严重，0.21的streaming方式无法正确使用combinefileinputformat，修改部分源码，以及实现CombineFileLineRecordReader。
源码修改部分：hadoop-mapred-0.21.0.jar包里的org.apache.hadoop.mapred.lib.CombineFileInputFormat.java文件

streaming方式要求任务分片为org.apache.hadoop.mapred.InputSplit，而实际输入分片为org.apache.hadoop.mapreduce.lib.input.CombineFileSplit，所以需要转变分片类型。

public InputSplit[] getSplits(JobConf job, int numSplits) 
    throws IOException {
	  List<org.apache.hadoop.mapreduce.InputSplit> splits = super.getSplits(new Job(job));
	  int size = splits.size();
	  
	  if (splits.get(0) instanceof org.apache.hadoop.mapreduce.lib.input.CombineFileSplit)
	  {
		  InputSplit[] returnSplits = new InputSplit[size];
	      for(int i=0;i<size; i++)
	      {
		      org.apache.hadoop.mapreduce.lib.input.CombineFileSplit combineFileSplit = (org.apache.hadoop.mapreduce.lib.input.CombineFileSplit)splits.get(i);
		      Path[] paths = combineFileSplit.getPaths();
		      long[] starts = combineFileSplit.getStartOffsets();
		      long[] lengths = combineFileSplit.getLengths();
		      String[] locations = combineFileSplit.getLocations();
		  
		      returnSplits[i] = new CombineFile