Track Grid requirements for distributed quota and intelligent overflow demos

[nerdalert]

Purpose

Track the Grid-side work needed to satisfy the two experimental demos:

• Praxis Grid - Distributed token rate limiting with Grid routing
• Praxis Grid - Intelligent Overflow

The token-rate-limit implementation itself belongs under the parent AI epic ai#121. This issue tracks the Grid integration, provider selection, admission, overlay, observability, deployment, and demo-validation work required around it.

Priority-ordered requirements

Priority	Requirement	Demo need	Existing issue(s)
1	Make metric-driven admission reliable: scrape queue metrics, withdraw saturated providers, preserve eligible providers, and recover when pressure clears.	Cloud burst must transition from local providers to overflow and back.	#10
2	Withdraw providers when health checks or gateway availability fail, while preserving the remaining routing candidates.	Both demos must avoid routing new requests to unavailable gateways.	#20, #11
3	Preserve locality and explicit provider-selection grouping across same-site, same-zone, same-region, and cross-region providers.	Distributed quota routing needs predictable provider groups; overflow must remain a separate fallback group.	#14
4	Publish complete, accepted overlay state and structured routing decisions, including provider identity, group, admission, weights, and revisions.	The UI and demo evidence must show why a request stayed local or burst.	#13, #19, #18
5	Keep Grid-generated consumer configuration compatible with the Praxis routing and quota pipeline.	Consumers must accept the live overlay and forward local and overflow requests correctly.	#36
6	Provide reproducible Helm deployment examples and lifecycle/teardown validation for the demo topology.	Both demos must be deployable from documented manifests without hand-applied state.	#15
7	Validate the complete demo topologies, including shared quota state across gateways, provider attribution, admission transitions, and recovery.	Establishes repeatable acceptance evidence for both demos.	#17
8	Align Grid architecture and deployment documentation with the actual routing and provider-selection contracts.	Users need enough information to reproduce the demos and understand control-plane versus request-path behavior.	#16, #1
9	Expose provider usage/spend signals where Grid is responsible for publishing them.	Intelligent Overflow needs provider-level cost attribution without inventing data.	#69

Acceptance relationship

• The distributed token-rate-limit demo proves authenticated, shared sliding-window quota enforcement across consumer gateways.
• The Intelligent Overflow demo proves Grid admission and provider grouping change the accepted overlay, while Praxis serves requests from that immutable snapshot.
• Grid must not perform a request-time control-plane lookup.
• Quota identity and usage must remain independent of provider group, provider gateway, or overflow destination.
• Existing issues above should be closed or explicitly marked as covered by the demos when their acceptance criteria are met.

Parent tracking

• Parent AI epic: praxis-proxy/ai#121
• Related AI implementation PR: praxis-proxy/ai#796

[nerdalert]

Superseded by #101. The replacement tracks the capability gaps after praxis-proxy/ai#796, keeps component ownership explicit, and treats the experimental demos as validation references rather than structuring the roadmap around them.