Handle 'lambda' and '@' in TIP theorems

[Chris Warburton]

See issue f5669cd4db415d36 of haskell-te

Some TIP theorems contain things like lambda functions and applications
'@' which aren't being handled properly.

We should add tests to check the behaviour of various keywords appearing
in theorem statements that we translate to JSON.

In the case of 'lambda', it's currently reasonable to abort (produce no
equation), since QuickSpec can't conjecture lambda functions anyway.

In the case of '@', if it's fully-applied, i.e. '(@ f x)', we should
simplify it away to get '(f x)', since in this case '@' is acting merely
as an annotation, to aid translation between TIP and languages which may
treat functions in some peculiar way (e.g. 2-lisps like Common Lisp,
where we might have to say '(funcall 'f x)' instead of just '(f x)' like
we can say in 1-lisps like Scheme).

I'm not sure if '@' appears without being fully-applied. Try grepping
the TIP benchmarks to see.