Fix alpha channel computation with ColorMatrixColorFilter

Bug #7222476

There were two issues:
- Blending was ignored with color filters
- The addition vector of a color filter was treated as integer values
  instead of float values

Change-Id: Id94065704a30ee8aaaa5724a9f3a3cff7c50ced7

libs/hwui/ProgramCache.cpp

@@ -347,7 +347,6 @@ const char* gFS_Main_ApplyColorOp[4] = {
         // None
         "",
         // Matrix
-        // TODO: Fix premultiplied alpha computations for color matrix
         "    fragColor *= colorMatrix;\n"
         "    fragColor += colorMatrixVector;\n"
         "    fragColor.rgb *= fragColor.a;\n",