Klever

This is quite an important bug that does not allow queueing jobs.

I hope that you will do it.

Merged in caf859214.

This is a quite complex feature. I have implemented a new resource manager that allows to track resources for many nodes and now it used in Native scheduler that has only one. Users now should set resource limits for tasks not in job.json but in tasks.json and this file must be always loaded to Bridge to start job execution. Good news is a possibility to set any limits without warning that deadlock will occur. So you can start any number of jobs with various priorities and go drinking coffee. Jobs that can be solved will be solved and inconsistent ones will be failed. No pending forever jobs, no deadlock are possible now. Open new issues if something will work wrong.