Don't include variable ids in inconsistent overload error messages

Summary:
I noticed while looking at the errors pyrefly generates on mypy_primer that we're printing variable ids in our error messages about inconsistent overloads. This causes error messages to vary between pyrefly runs.

Fixed by making sure that when we substitute fresh variables into a callable signature, we use the pre-substitution signature for error messages.

Reviewed By: ndmitchell

Differential Revision: D81898793

fbshipit-source-id: fd6c99a81525a5ba6c5510efe9905787e106baf2

Author: rchen152

pyrefly/lib/alt/function.rs

@@ -1109,7 +1109,7 @@ impl<'a, Ans: LookupAnswer> AnswersSolver<'a, Ans> {
                 &|| {
                     TypeCheckContext::of_kind(TypeCheckKind::OverloadInput(
                         overload_func.signature.clone(),
-                        impl_func.signature.clone(),
+                        impl_sig.clone(),
                     ))
                 },
             );

pyrefly/lib/test/overload.rs

@@ -688,3 +688,23 @@ def foo(d: TD | int) -> None: ...
 assert_type(foo({ "x": "foo" }), Any) # E: No matching overload found for function `foo`
     "#,
 );
+
+testcase!(
+    test_generic_impl_input_inconsistent,
+    r#"
+from typing import overload, TypeVar
+T = TypeVar("T")
+S = TypeVar("S")
+
+# The implementation signature is intentionally inconsistent with this overload signature
+# (`exception` is kw-only in the implementation) so that we can test how legacy TypeVars are
+# printed in the error message.
+@overload
+def catch(exception: T) -> T: ...  # E: Implementation signature `(f: S | None = None, *, exception: T) -> S | T` does not accept all arguments that overload signature `(exception: object) -> object` accepts
+
+@overload
+def catch(f: S, *, exception: T) -> S | T: ...
+
+def catch(f: S | None = None, *, exception: T) -> S | T: ...
+    "#,
+);