fix(BA-2404): Generalize gpu_allocated field with improved SlotName and ResourceSlot typing

[achimnol]

resolves #5926 (BA-2404)

Core changes:

• Rewrite SlotName to be a UserString subclass which lazily parses the slot name format.
  • {device_name}.{major_type}[:{minor_type}]
• Update gpu_allocated calculation to use SlotName.is_accelerator() method to sum all allocated accelerators
  • Centralize the logic to guess if the slot name is an accelerator or not, for the future.
• Update ResourceSlot to use UserDict with proper generic type arguments
  • Since typeshed and the stdlib documentation indicates initialization should accept a single dict instance rather than kwargs, update all of such usage.

Warning

When there are multiple different accelerators allocated in a single container or installed in a single agent node, this simple "sum" for gpu_allocated may have a non-sense value when mixing different units and fraction scales. We need some design discussion here.

Checklist: (if applicable)

• Milestone metadata specifying the target backport version
• Mention to the original issue
• Test case(s) to:
  • Demonstrate the difference of before/after
  • Demonstrate the flow of abstract/conceptual models with a concrete implementation