Changes:
    ========

    - Allows launching kernels locally, if None is specified as a
      parameters list for `salloc` in Jupyter config. This behaviour is
      needed to be enabled in Jupyter config with a parameter:
        c.SlurmKernelManager.default_kernel_local = False

    - Allows fallback to local kernel, if SLURM shell has failed to
      spawn. This behaviour is needed to be enabled in Jupyter config
      with a parameter:
        c.SlurmKernelManager.fallback_kernel_local = False

    - (Hopefully) better stack traces, if something goes wrong.

    1) There was a branch with an attempt to execute kernel in None
       shell.
    2) There was a variant, when kernel would not fallback to local,
       even if it is specified in config.

    1) Cleaner exceptions.
    2) Force full restart even for local kernels.

    1) six.reraise instead of six.raise_from
    2) Correct handling of default launch parameters, if local kernel is
       not allowed.

    Sending signals to kernel group properly using `scancel`.

    Since we can't guarantee, that new kernel node will be the same
    as old node, we have to request the full shutdown and full
    regeneration of a connection file.

    Random ports cleanup function are pulled from trunk version of
    jupyter_client and will be removed after they appear in stable
    versions.

    Trying more to follow logic of a reference realization

    Checking scontrol output and returning True, if status of the kernel
    job is found to be `JobState=RUNNING`.

    Able to launch a kernel now