Decorating anonymous functions is inconsistent

[allison-casey]

Attempting to decorate an fn that can compile down to a lambda fails, but it won't fail if the fn contains statements that force compilation to an anonymous def

=> (with-decorator cache (fn [x] x))
Traceback (most recent call last):
  File "stdin-ecb69bbf92e18e5e87b4bc8c6fd3222c5969d4a8", line 1, in <module>
    (with-decorator cache (fn [x] x))
  File "<stdin>", line 1
    (with-decorator cache (fn [x] x))
                          ^
hy.errors.HySyntaxError: Decorated a non-function

=> (with-decorator cache (fn [x] (print 1) x))
@cache
def _hy_anon_var_2(x):
    print(1)
    return x
_hy_anon_var_2

I think it makes the most sense to force the anon def when an fn uses a decorator like it already does when it contains statements or type annotations.

[Kodiologist]

This should be immediately closed by #2270, since there will no longer be a way to decorate a fn.

[allison-casey]

then it should be noted in NEWS as a breaking change probably

[Kodiologist]

I think that's implied by "with-decorator and #@ have been removed in favor of decorator lists (see below)" and the description of decorator lists only mentioning defn and defclass.