Autocorrection for Lint/UnusedMethodArgument should update Sorbet sig

[dduugg]

---

Expected behavior

Per #11020 (comment) the autocorrect for Lint/UnusedMethodArgument should update the Sorbet type signature, e.g.

sig { params(x: Integer).void }
def foo(x); end

should autocorrect to:

sig { params(_x: Integer).void }
def foo(_x); end

(Alternatively, the cop could be marked SafeAutoCorrect: false, which was attempted in #11020 )

Actual behavior

The parameter is renamed under autocorrection, but not the type signature:

sig { params(x: Integer).void }
def foo(_x); end

This code will fail to typecheck, as well as resulting in a RuntimeError on method invocation.

Steps to reproduce the problem

run rubocop -a on this snippet:

sig { params(x: Integer).void }
def foo(x); end

RuboCop version

Include the output of rubocop -V or bundle exec rubocop -V if using Bundler.
If you see extension cop versions (e.g. rubocop-performance, rubocop-rspec, and others)
output by rubocop -V, include them as well. Here's an example:

$ bundle exec rubocop -V
1.36.0 (using Parser 3.1.2.1, rubocop-ast 1.21.0, running on ruby 3.1.1) [x86_64-darwin21]

[dduugg]

On further consideration, the autocorrect would still break rbi/rbs files, and I'm not sure rubocop supports the necessary correction across files (and file tops). I think the cop still needs to be marked SafeAutoCorrect: false. What do you think, @koic ?