Add and use a separate kylin_job_conf.xml for in-mem cubing

From v1.5.0, kylin has two cubing algorithm: by-layer and by-split(in-mem).

The in-mem cubing require more memory resource for the mapper; so far it is configured by parameter "kylin.job.cubing.inmem.mrjob_conf_override" with default value "mapreduce.map.java.opts=-Xmx2700m; mapreduce.map.memory.mb=3072; mapreduce.task.io.sort.mb=200"; This is a little confusing as user can also configure such parameters in kylin_job_conf.xml

A better way is adding a new config file for in-mem cubing job, moving the in-mem specific mapreduce parameters there.