Add support for fair-share scheduling to be based upon resource use at the level of bank accounts and ignore use of individual users. Patch by Par Andersson, National Supercomputer Centre, Sweden.

- The node count in the bluegene select plugin job record was getting set to 
    zero, which prevented the job from re-running after a requeue
- Fix bug in new kill job logic which tried to set the job's start time in 
   the RPC from slurmctld to slurmd. We need to leave the value as zero rather 
   than using a NULL job pointer if the job is not found

before searching for keywords

    --nodes option was ignored).

Added descriptions of several new contributed files/directories

hostlist: factor out a pattern

This is a clerical patch which transforms a repeating pattern of type

 	if (dims > 1) {
		if (hr->width == dims) {
			// Alternative 'A'
		} else {
			// Alternative 'B'
		}
	} else {
		// Alternative 'B'
	}

into the shorter pattern

	if (dims > 1 && hr->width == dims) {
		// Alternative 'A'
	} else {
		// Alternative 'B'
	}

Miscellaneous typos, found while reading documentation.

The majority is in the manpages and within comments, only one occurrence
is in an error message (function name of set_node_down).

Allow compilation with --enable-cray.

This patch allows compilation of slurm with --enable-cray without build errors,
on such (multi-cluster) systems that need to be able to load select/cray, but
without actually using the internal functionality (e.g. slurmdbd host).

The code stubs have been translated into corresponding comments with FIXMEs
added, these are then addressed later in the (post-2.2) patch set.

Added the missing part (variables) to fix build errors and tested tested that 
the compilation of --enable-cray works without errors.