Wingman should notice variables already used, outside of the hole

[isovector]

test :: (a -> b -> b) -> b -> [a] -> b
test _ b [] = b
test f_b b (a : l_a4) = f_b a $ test f_b b _

Wingman wants to fill this hole with a : l_a4, since it uses more variables from the hypothesis. But it doesn't pay attention to the expression context, where it could learn that a has already been used.

This is not hard to fix --- subtract used-in-context variables from syn_used_vals before scoring.

[isovector]

Thinking further about this, we can use the provenance of a term to calculate its "specificity." If the user has gone through all of this work to get a specific variable in scope, which happens to have the same type as one more widely available, then it's probable that we should use the more specific one. At least for non-top holes.