HdfsWriter delete wrong files when execute on Windows. Windows机器下执行HDFS写入存在删数据的问题。

在Windows机器下写入到HDFS时，有严重bug，会导致指定path下的所有数据删除，附上同一个json配置在windows和Linux下执行的日志：

### windows下的DataX执行日志：
2018-06-21 18:08:52.665 [job-0] INFO HdfsWriter$Job - start delete tmp dir [hdfs://172.30.120.15:8020/inceptor1/user/hive/warehouse/default.db/hive] .
2018-06-21 18:08:52.761 [job-0] INFO HdfsWriter$Job - finish delete tmp dir [hdfs://172.30.120.15:8020/inceptor1/user/hive/warehouse/default.db/hive] .

### linux下的DataX执行日志：
2018-06-21 16:42:43.633 [job-0] INFO HdfsWriter$Job - start delete tmp dir [hdfs://172.30.120.15:8020/inceptor1/user/hive/warehouse/default.db/hive/test__b4800119_af56_4cf2_b4e4_af49751f69ac] .
2018-06-21 16:42:43.642 [job-0] INFO HdfsWriter$Job - finish delete tmp dir [hdfs://172.30.120.15:8020/inceptor1/user/hive/warehouse/default.db/hive/test__b4800119_af56_4cf2_b4e4_af49751f69ac] .

看了下源码，是在设置文件路径分隔符时，选用的是DataX运行环境的文件分隔符，而不是HDFS运行环境的文件分隔符。
源码路径：com/alibaba/datax/plugin/writer/hdfswriter/HdfsWriter.java

Start by reading com/alibaba/datax/plugin/writer/hdfswriter/HdfsWriter.java and tracing how the temporary HDFS deletion path is assembled on Windows versus Linux. Compare the logged paths with the configured target; done means Windows deletes only the generated temporary directory, matching the Linux behavior, rather than the configured base path.

Written by the indexing model from the issue text.