This feature tacks the design implementation and delivery of a Resource Provider weigher to support resouce aware schdluling based on un-requested capacity and provider traits

This feature proposes a new scheduler weigher that analyzes resource
provider tree data from Placement to make more intelligent host
selection decisions. Unlike existing weighers that focus on single
resource types (RAM, disk), the ResourceProviderWeigher considers
multiple resource classes and traits simultaneously.

The weigher implements a "most boring host" strategy, preferring
hosts that closely match instance requirements without unnecessary
specialized capabilities. This optimizes resource utilization by
reserving specialized resources (GPUs, SR-IOV, high-memory nodes)
for workloads that actually need them.

Key features:

• Processes existing provider summaries from Placement (no new API
  calls)
• Calculates scores based on resource utilization and trait matching
• Disabled by default (weight multiplier = 0.0) for backward
  compatibility
• Configurable to limit which resources and traits are considered

This work was scoped down from a larger CPU performance tiering
proposal to focus specifically on the weigher component, which can
be used independently for various resource management scenarios.

https://review.opendev.org/c/openstack/nova-specs/+/951222