Clarify optimizeimg docs; add warnings

The validator will now warn if it detects that a crushed output
is fed into something that is not a crusher.

The is_crusher method of an optimizer shall return True if the
optimisation process is lossless, and does try to find optimal
encoding parameters as opposed to only removing unneeded channels
or reducing palettes.

overviewer_core/optimizeimages.py

@@ -39,6 +39,10 @@ class Optimizer:
 
         if (not exists_in_path(self.binaryname)) and (not exists_in_path(self.binaryname + ".exe")):
             raise Exception("Optimization program '%s' was not found!" % self.binaryname)
+    
+    def is_crusher(self):
+        """Should return True if the optimization is lossless, i.e. none of the actual image data will be changed."""
+        raise NotImplementedError("I'm so abstract I can't even say whether I'm a crusher.")
         
 
 class NonAtomicOptimizer(Optimizer):
@@ -84,6 +88,9 @@ class pngnq(NonAtomicOptimizer, PNGOptimizer):
 
         NonAtomicOptimizer.fire_and_forget(self, args, img)
 
+    def is_crusher(self):
+        return False
+
 class pngcrush(NonAtomicOptimizer, PNGOptimizer):
     binaryname = "pngcrush"
     # really can't be bothered to add some interface for all
@@ -98,6 +105,9 @@ class pngcrush(NonAtomicOptimizer, PNGOptimizer):
 
         NonAtomicOptimizer.fire_and_forget(self, args, img)
 
+    def is_crusher(self):
+        return True
+
 class optipng(Optimizer, PNGOptimizer):
     binaryname = "optipng"
 
@@ -106,6 +116,9 @@ class optipng(Optimizer, PNGOptimizer):
     
     def optimize(self, img):
         Optimizer.fire_and_forget(self, [self.binaryname, "-o" + str(self.olevel), "-quiet", img])
+
+    def is_crusher(self):
+        return True
         
 
 def optimize_image(imgpath, imgformat, optimizers):