WIP: Fix coretime partioning + super low latency on-demand

[eskimor]

TODOS:

• Make it typecheck
• Make CoreState one state item for all cores
• Runtime APIs must look into the future (retrieve claim queue for now +1)
• Update scheduler + paras_inherent to use the new pop all cores at once API
• Fix scheduler and paras_inherent to use the new API
• Add back duplication of first assignment for on-demand
• Limit assignments returned by the scheduler to availability cores count
• Fix peek for on-demand
• Fix benchmarks
• Migrations
• Add test that peek at block x correctly predicts pops at block x + n
• Add test that on-demand order on empty core shows up in two next blocks (and no more)
• Add test that bulk assignment coming in on an unoccupied core gets duplicated before the start
• Add test that peek does not modify state (multiple calls will yield the same result)
• Fix tests
• Update docs
• Drop other parachain assigner
• Move coretime & ondemand assigner into scheduler folder
• Migration OnDemandAssignmentProvider -> OnDemand (Westend)
• Polish

Fixes: #1312 #6563

[paritytech-workflow-stopper]

All GitHub workflows were cancelled due to failure one of the required jobs.
Failed workflow url: https://github.com/paritytech/polkadot-sdk/actions/runs/19176567296
Failed job name: build-rustdoc