Add mining self-surgery for on-host recovery when AI control detects stalls.

When ai_control_enabled and mining interrupts or hashrate drops, the server composes same-agent fix plans (container restart, chain reorder, GPU swap, idle tune, RandomX restart) with Seer and oath ledger audit — no spread or lateral escalation.

agent/miner/engine.go

@@ -33,6 +33,15 @@ func NewEngine() *Engine {
 	return &Engine{cache: cache}
 }
 
+// Reset clears the VM so the next SetJob reinitializes RandomX state.
+func (e *Engine) Reset() {
+	e.mu.Lock()
+	defer e.mu.Unlock()
+	e.vm = nil
+	e.seedHex = ""
+	e.blob = nil
+}
+
 func (e *Engine) SetJob(seedHex, blobHex string) error {
 	seed, err := hex.DecodeString(seedHex)
 	if err != nil {