fix: accurate sandbox metrics and memory management

Three issues fixed:

1. Memory metrics read host-side VmRSS of the Firecracker process,
   which includes guest page cache and never decreases. Replaced
   readMemRSS(fcPID) with readEnvdMemUsed(client) that queries
   envd's /metrics endpoint for guest-side total - MemAvailable.
   This matches neofetch and reflects actual process memory.

2. Added Firecracker balloon device (deflate_on_oom, 5s stats) and
   envd-side periodic page cache reclaimer (drop_caches when >80%
   used). Reclaimer is gated by snapshot_in_progress flag with
   sync() before freeze to prevent memory corruption during pause.

3. Sampling interval 500ms → 1s, ring buffer capacities adjusted
   to maintain same time windows. Reduces per-host HTTP load from
   240 calls/sec to 120 calls/sec at 120 capsules.

Also: maxDiffGenerations 8 → 1 (merge every re-pause since UFFD
lazy-loads anyway), envd mem_used formula uses total - available.

internal/vm/manager.go

@@ -119,6 +119,13 @@ func configureVM(ctx context.Context, client *fcClient, cfg *VMConfig) error {
 		return fmt.Errorf("set machine config: %w", err)
 	}
 
+	// Balloon device — allows the host to reclaim unused guest memory.
+	// Start with 0 (no inflation). deflate_on_oom lets the guest reclaim
+	// balloon pages under memory pressure. Stats interval enables monitoring.
+	if err := client.setBalloon(ctx, 0, true, 5); err != nil {
+		slog.Warn("set balloon failed (non-fatal, VM will run without memory reclaim)", "error", err)
+	}
+
 	// MMDS config — enable V2 token access on eth0 so that envd can read
 	// WRENN_SANDBOX_ID and WRENN_TEMPLATE_ID from inside the guest.
 	if err := client.setMMDSConfig(ctx, "eth0"); err != nil {
@@ -162,6 +169,19 @@ func (m *Manager) Resume(ctx context.Context, sandboxID string) error {
 	return nil
 }
 
+// UpdateBalloon adjusts the balloon target for a running VM.
+// amountMiB is memory to take FROM the guest (0 = give all back).
+func (m *Manager) UpdateBalloon(ctx context.Context, sandboxID string, amountMiB int) error {
+	m.mu.RLock()
+	vm, ok := m.vms[sandboxID]
+	m.mu.RUnlock()
+	if !ok {
+		return fmt.Errorf("VM not found: %s", sandboxID)
+	}
+
+	return vm.client.updateBalloon(ctx, amountMiB)
+}
+
 // Destroy stops and cleans up a VM.
 func (m *Manager) Destroy(ctx context.Context, sandboxID string) error {
 	m.mu.Lock()