Roihu updates: Batch queue system

[rkronberg]

• There will be quite many queues accounting for old use cases from Puhti & Mahti. Important to highlight pitfalls of using node-based partitions when code is serial, for example.
• Maybe also remind here about the special resources like (disaggregated) NVMe
• You can only submit jobs to GPU partition from GPU login & CPU partition form CPU login
• Otherwise, fundamentally, Slurm will work as before. Small changes like partition names etc.