CLAUDE: Update filterConv5dROI to use vel_suppr_rad neighbourhood

Replace fixed ±1 pixel/velocity loops and indx_center_3d3/indx_margins_3d3
with configurable vel_suppr_rad loops and indx_center_vsupp/indx_margins_vsupp.
Co-authored-by: Claude <claude@elphel.com>

src/main/java/com/elphel/imagej/cuas/rt/CuasRTUtils.java

@@ -1506,13 +1506,13 @@ public class CuasRTUtils {
 
     /**
      * 4D non-maximum suppression on a ROI fine-velocity result.
-     * For each (pixel, velocity) point, find the maximum M over the ±1 neighbourhood
-     * in both pixel space (3×3) and velocity space (3×3 in vx/vy), then:
-     *   if M >= athresh  AND  val < rthresh * M  →  zero the point (sidelobe suppression)
+     * For each (pixel, velocity) point, find the maximum M over the ±vel_suppr_rad neighbourhood
+     * in both pixel space and velocity space (vx/vy), then:
+     *   thresh = max(athresh, M*rthresh); suppress if (M - val) >= thresh
      *   otherwise keep val unchanged (including the weak-signal passthrough when M < athresh).
      *
-     * Uses indx_center_3d3 / indx_margins_3d3 for bit-reversed multithreaded iteration
-     * (same kernel radius = 1 as the pixel neighbourhood here).
+     * Uses indx_center_vsupp / indx_margins_vsupp for bit-reversed multithreaded iteration
+     * (kernel radius = vel_suppr_rad matches both pixel and velocity neighbourhood).
      *
      * @param data     [roi_npix][nsub][nvel] — ROI-indexed output of convolve3d(roi)
      * @param roi      pixel rectangle (full-image coordinates)
@@ -1537,8 +1537,8 @@ public class CuasRTUtils {
     	// Center pixels: all 8 pixel neighbours are within image bounds; only ROI check needed.
     	for (int ithread = 0; ithread < threads.length; ithread++) {
     		threads[ithread] = new Thread() { public void run() {
-    			for (int nPix = ai.getAndIncrement(); nPix < indx_center_3d3.length; nPix = ai.getAndIncrement()) {
-    				int ipix = indx_center_3d3[nPix];
+    			for (int nPix = ai.getAndIncrement(); nPix < indx_center_vsupp.length; nPix = ai.getAndIncrement()) {
+    				int ipix = indx_center_vsupp[nPix];
     				int px = ipix % width, py = ipix / width;
     				if (px < roi_x0 || px >= roi_x1 || py < roi_y0 || py >= roi_y1) continue;
     				int roi_pix = (py - roi_y0) * roi.width + (px - roi_x0);
@@ -1547,16 +1547,16 @@ public class CuasRTUtils {
     						double val = data[roi_pix][sub_idx][v_out_idx];
     						int vy = v_out_idx / vout_dim, vx = v_out_idx % vout_dim;
     						double M = 0;
-    						for (int dpy = -1; dpy <= 1; dpy++) {
+    						for (int dpy = -vel_suppr_rad; dpy <= vel_suppr_rad; dpy++) {
     							int ny = py + dpy;
-    							for (int dpx = -1; dpx <= 1; dpx++) {
+    							for (int dpx = -vel_suppr_rad; dpx <= vel_suppr_rad; dpx++) {
     								int nx = px + dpx;
     								if (nx < roi_x0 || nx >= roi_x1 || ny < roi_y0 || ny >= roi_y1) continue;
     								int nroi = (ny - roi_y0) * roi.width + (nx - roi_x0);
-    								for (int dvy = -1; dvy <= 1; dvy++) {
+    								for (int dvy = -vel_suppr_rad; dvy <= vel_suppr_rad; dvy++) {
     									int nvy = vy + dvy;
     									if (nvy < 0 || nvy >= vout_dim) continue;
-    									for (int dvx = -1; dvx <= 1; dvx++) {
+    									for (int dvx = -vel_suppr_rad; dvx <= vel_suppr_rad; dvx++) {
     										int nvx = vx + dvx;
     										if (nvx < 0 || nvx >= vout_dim) continue;
     										double d = data[nroi][sub_idx][nvy * vout_dim + nvx];
@@ -1578,8 +1578,8 @@ public class CuasRTUtils {
     	ai.set(0);
     	for (int ithread = 0; ithread < threads.length; ithread++) {
     		threads[ithread] = new Thread() { public void run() {
-    			for (int nPix = ai.getAndIncrement(); nPix < indx_margins_3d3.length; nPix = ai.getAndIncrement()) {
-    				int ipix = indx_margins_3d3[nPix];
+    			for (int nPix = ai.getAndIncrement(); nPix < indx_margins_vsupp.length; nPix = ai.getAndIncrement()) {
+    				int ipix = indx_margins_vsupp[nPix];
     				int px = ipix % width, py = ipix / width;
     				if (px < roi_x0 || px >= roi_x1 || py < roi_y0 || py >= roi_y1) continue;
     				int roi_pix = (py - roi_y0) * roi.width + (px - roi_x0);
@@ -1588,18 +1588,18 @@ public class CuasRTUtils {
     						double val = data[roi_pix][sub_idx][v_out_idx];
     						int vy = v_out_idx / vout_dim, vx = v_out_idx % vout_dim;
     						double M = 0;
-    						for (int dpy = -1; dpy <= 1; dpy++) {
+    						for (int dpy = -vel_suppr_rad; dpy <= vel_suppr_rad; dpy++) {
     							int ny = py + dpy;
     							if (ny < 0 || ny >= height) continue;
-    							for (int dpx = -1; dpx <= 1; dpx++) {
+    							for (int dpx = -vel_suppr_rad; dpx <= vel_suppr_rad; dpx++) {
     								int nx = px + dpx;
     								if (nx < 0 || nx >= width) continue;
     								if (nx < roi_x0 || nx >= roi_x1 || ny < roi_y0 || ny >= roi_y1) continue;
     								int nroi = (ny - roi_y0) * roi.width + (nx - roi_x0);
-    								for (int dvy = -1; dvy <= 1; dvy++) {
+    								for (int dvy = -vel_suppr_rad; dvy <= vel_suppr_rad; dvy++) {
     									int nvy = vy + dvy;
     									if (nvy < 0 || nvy >= vout_dim) continue;
-    									for (int dvx = -1; dvx <= 1; dvx++) {
+    									for (int dvx = -vel_suppr_rad; dvx <= vel_suppr_rad; dvx++) {
     										int nvx = vx + dvx;
     										if (nvx < 0 || nvx >= vout_dim) continue;
     										double d = data[nroi][sub_idx][nvy * vout_dim + nvx];