Follow-up from #785

The Providers page (#785, PR #893) shipped health metrics (calls/24h, response time, error rate) but two scope items need backend model changes:

1. Usage metrics: total tokens and cost per provider

Issue #785 specifies "calls in last 24h, total tokens, cost per provider" as usage metrics. Only calls/24h is implemented.

Required changes:

• Add total_tokens_24h and total_cost_24h fields to ProviderHealthSummary (backend)
• Wire these into the health tracking aggregation pipeline
• Display in ProviderHealthMetrics.tsx and ProviderCard.tsx (frontend)

2. Model capabilities display

Issue #785 specifies "available models per provider with capabilities". Models are listed with ID, alias, context, and pricing -- but capabilities (vision, tools, streaming) are not shown.

Required changes:

• Add capabilities fields to ProviderModelConfig (or surface from ModelCapabilities in the driver layer)
• Expose in the GET /providers/{name}/models response
• Display capability badges in ProviderModelList.tsx (frontend)

Labels

• Parent: feat: Providers page (health status, usage metrics, model listing) #785