feat(api): expose model_source filter in dashboard endpoints

Add model_source query parameter to GetModelStats and GetUserBreakdown handlers with explicit IsValidModelSource validation. Include model_source in cache key to prevent cross-source cache hits. Expose upstream_model in usage log DTO with omitempty semantics.

backend/internal/handler/dto/mappers.go

@@ -521,6 +521,7 @@ func usageLogFromServiceUser(l *service.UsageLog) UsageLog {
 		AccountID:             l.AccountID,
 		RequestID:             l.RequestID,
 		Model:                 l.Model,
+		UpstreamModel:         l.UpstreamModel,
 		ServiceTier:           l.ServiceTier,
 		ReasoningEffort:       l.ReasoningEffort,
 		InboundEndpoint:       l.InboundEndpoint,

backend/internal/handler/dto/types.go

@@ -332,6 +332,9 @@ type UsageLog struct {
 	AccountID int64  `json:"account_id"`
 	RequestID string `json:"request_id"`
 	Model     string `json:"model"`
+	// UpstreamModel is the actual model sent to the upstream provider after mapping.
+	// Omitted when no mapping was applied (requested model was used as-is).
+	UpstreamModel *string `json:"upstream_model,omitempty"`
 	// ServiceTier records the OpenAI service tier used for billing, e.g. "priority" / "flex".
 	ServiceTier *string `json:"service_tier,omitempty"`
 	// ReasoningEffort is the request's reasoning effort level.