Use max_count, not available_count, for detemrining partition sizes

As well, since we will now often have more nodes defined in Slurm than could be started, we should add a monitor process that will keep the number of potential nodes that could be allocated in sync with the availability according to CycleCloud.

Start by finding the partition-size calculation that chooses between max_count and available_count, then trace how CycleCloud reports node availability. Done means partition sizing uses the intended count and a monitor keeps potential allocations synchronized; the issue names no file or test, so identify relevant coverage while reading.

Written by the indexing model from the issue text.