node, process and thread count for SGE

[sverhoeven]

Figure out how to map counts to SGE

Refs #625

[sverhoeven]

The pe allocation_rule field (https://man.cx/sge_pe(5)) determines how slots are divided over the available hosts.

Determining which pe should be used when user request

1. single node and a single core per node ->
   1. no pe needed
2. single node and multiple cores ->
   1. Ignore pe with allocation_rule=$fill_up or $round_robin as those can not guarantee all slots will be on same host.
   2. Maybe use pe with allocation_rule=$pe_slots, but if requested core count is greater than any host (in the queue) then job will stay queued forever.
   3. Use pe with allocation_rule == number of requested cores
   4. Maybe use pe with allocation_rule > number of requested cores, will claim more slots than requested
3. multiple nodes and muliple cores per node ->
   1. Ignore pe with allocation_rule=$fill_up or $round_robin as those can not guarantee all slots will be on
      spread evenly among hosts.
   2. Ignore pe with allocation_rule=$pe_slots as that is for single host only
   3. Use pe with allocation_rule == number of requested cores, should use (number of nodes * core per node) as value for -pe
   4. Ignore pe with allocation_rule > number of requested cores, as that does not guarantee all slots will be spread evenly among hosts

When no valid pe can be found an exception should be thrown which tells user to use addSchedulerArgument("-pe", "", "") if they know better.

[sverhoeven]

When queue name is given then must check that pe is allowed in queue.

When queue name is not given then GE will pick a queue which has the requested pe so the best we can do is to check if there is any queue that has that pe

[sverhoeven]

Fixes in #631