feat: correx-managed model lifecycle slice 5 — Go TUI model swap + telemetry

Go TUI decodes model.changed / model.list / resource.status and renders a
status-bar VRAM/GPU%/RAM gauge plus a models overlay (m key / palette 'models'):
lists configured models with the resident one marked, enter swaps (SwapModel),
c clears the pin (ClearModelPin). All three new server messages are
non-event-bearing control/gauge frames, applied immediately like
provider.status_changed.

Server addition required for the picker: ServerMessage.ModelList (model.list,
NonEventMessage) sent once in the initial snapshot from
ManagedInferenceRouter.availableModelIds()/currentModelId() — the TUI otherwise
cannot enumerate swap targets.

Completes the 5-slice model-lifecycle feature. ./gradlew check green; go build/vet clean.

Plan: docs/plans/2026-05-31-model-lifecycle-management.md (slice 5 of 5).

apps/tui-go/internal/protocol/protocol.go

@@ -38,6 +38,9 @@ const (
 	TypeSnapshotComplete = "snapshot_complete"
 	TypeWorkflowList     = "workflow.list"
 	TypeToolAssessed     = "tool.assessed"
+	TypeModelChanged     = "model.changed"
+	TypeModelList        = "model.list"
+	TypeResourceStatus   = "resource.status"
 )
 
 // ServerMessage is a flat decode of every server->client variant. Field names
@@ -69,6 +72,20 @@ type ServerMessage struct {
 	ElapsedMs       int64 `json:"elapsedMs"`
 	LastSequence    int64 `json:"lastSequence"`
 
+	// model.changed
+	ModelID string `json:"modelId"`
+	Loaded  bool   `json:"loaded"`
+
+	// model.list (current reuses the field below)
+	Models  []string `json:"models"`
+	Current string   `json:"current"`
+
+	// resource.status (nil = unavailable)
+	GpuMemoryUsedMb   *int64 `json:"gpuMemoryUsedMb"`
+	GpuMemoryTotalMb  *int64 `json:"gpuMemoryTotalMb"`
+	GpuUtilizationPct *int   `json:"gpuUtilizationPct"`
+	ProcessRssMb      *int64 `json:"processRssMb"`
+
 	State          *SessionStateDto   `json:"state"`
 	RiskSummary    *RiskSummaryDto    `json:"riskSummary"`
 	Status         *ProviderHealthDto `json:"status"`
@@ -206,6 +223,16 @@ func Ping(ts int64) []byte {
 	return encode("Ping", map[string]any{"timestamp": ts})
 }
 
+// SwapModel asks the server to make modelID resident and pin it.
+func SwapModel(modelID string) []byte {
+	return encode("SwapModel", map[string]any{"modelId": modelID})
+}
+
+// ClearModelPin releases the manual-swap pin so per-stage selection resumes.
+func ClearModelPin() []byte {
+	return encode("ClearModelPin", map[string]any{})
+}
+
 // CreateGrant pre-authorizes a tool/tier so future calls skip the approval gate.
 // scope is "SESSION" or "STAGE"; permittedTiers are tier names like "T3".
 func CreateGrant(sessionID, scope string, permittedTiers []string, reason, toolName string) []byte {