Clear windows branch; preserve previous contents in win-old

kernels/k/dltss_pwin_dec0_layer.hip

@@ -1,10 +0,0 @@
-// CU mode: the waves of a window share one L0 cache (1-6% faster here; slower in the wide layers)
-// d4r-build-flags: -mcumode
-#define PWIN_F32ACC
-#ifndef D4R_K_POS_MT
-#define D4R_K_POS_MT 2
-#elif !defined(__GFX12__) || !defined(D4R_K_FP16_BASELINE)
-#error "experimental position token tiling requires gfx12 strict K baseline"
-#endif
-#include "pwin_pos.h"
-POS_DECODER(dltss_pwin_dec0_layer, 2, 32, 64, 40, 48, D4R_K_POS_MT)