Use 'F' as sentinel token

In implementing the random rule generation, I had mistakenly believed
that 'f' meant 'step and draw' while 'F' meant only 'step'. In reality
the opposite is true.

The consequence of this mistake is that most generated rules will result
in pictures with gaps.

This is a partial fix for #83

tools/random-production-rules.py

@@ -147,7 +147,7 @@ def generate_lhs_tokens(distribution, rng):
     tokens = []
     token_pool = list(distribution.keys())
     token_probabilities = list(distribution.values())
-    while "f" not in tokens:
+    while "F" not in tokens:
         tokens.append(rng.choice(token_pool, p=token_probabilities))
 
     return tokens