shave some cycles (save/restore Y instead of recomputing)

src/fx/fx.dhgr.precomputed.1bit.a

@@ -77,12 +77,12 @@ input=*+1
          +ROW_X_TO_MIRROR_ADDRESSES
 
          inc   input
-         clc
-         !byte $2C
-bankloop sec
-         tay
          lda   (input), y
          +HIGH_3_LOW_5 input
+         sty   <y
+         clc
+         !byte $A9
+bankloop sec
 
          ; main 1x2 block in top-left quadrant
 src1=*+1
@@ -128,7 +128,8 @@ mirror_dest2=*+1
          +COPY_BIT mirror_src1, mirror_dest1, copymasks
          +COPY_BIT mirror_src2, mirror_dest2, copymasks
 
-         lda   #0
+y=*+1
+         ldy   #$FD
          sta   $C003
          sta   $C005
          bcc   bankloop