feedback for the `strict_top_level_inference` lint

[devoncarew]

This is general feedback for the strict_top_level_inference lint; cc @srawlins.

In order to try this out, I made a custom version of package:lints/core.yaml and ran it on the sdk repo. Some notes:

• it found 24,835 issues
• 24,374 of them were in test code (so, all, but 100 issues)
• from spot checking them, many of them involved a test method that did not have a type annotation (so, tests not as closure but as individual methods; test_functionTypedFormalParameter() async { ... })
• for these cases it would be useful to have a quick fix that added a return type (for many of these cases, this would be void, or Future, or Future<void>).

I think all the cases it fired on were legit. My main feedback from above is that a quick fix would be useful :)

[srawlins]

Ah yeah, I noted in the CL that we cannot predict what the missing type should be. But I was only thinking of field types and parameter types.

For return types, we should definitely be able to recommend a type. Maybe not in all cases, but the void and Future<void> ones should be very easy (no return statements? void or Future<void>!).

[devoncarew]

Yeah, I don't think offering to quick-fix the return type to dynamic would be helpful, but if we can determine the right type (void, ...), that would be great.

Perhaps the impl. of the quick fix for always_declare_return_types would be helpful here?

[srawlins]

I forgot to link the CR, but https://dart-review.googlesource.com/c/sdk/+/401800 links the quick fix that you suggest.

I don't think that this quick fix ever suggests a return type when the function body has no return statements (it would be either void or Future<void>). I can add that too.