Infer attr_accessor writer signatures

[Morriar]

Summary

This PR teaches gem RBI generation to infer signatures for writers generated by attr_accessor.

When code uses:

sig { returns(Integer) }
attr_accessor :foo

sorbet-runtime attaches the pending signature to the generated reader (foo) only. The generated writer (foo=) has no runtime signature, so Tapioca previously emitted it as untyped:

sig { returns(::Integer) }
def foo; end

def foo=(_arg0); end

With this change, Tapioca infers the writer signature from the matching reader signature:

sig { returns(::Integer) }
def foo; end

sig { params(foo: ::Integer).returns(::Integer) }
def foo=(foo); end

Details

The inference is intentionally conservative:

• It only applies when the writer has no runtime signature.
• The method must look like an attr-style writer:
  • method name ends with =
  • exactly one unnamed required parameter
• A matching reader must exist on the same constant.
• The reader must have a simple zero-argument signature.
• The writer and reader’s original source locations must match, ensuring we only infer signatures for methods generated by the same attr_accessor call and avoid separately-defined or manually-defined writer methods.

[Morriar]

Handled the review comments:

• inferred writer signatures now return the reader type instead of void
• moved the source-location check into reader matching
• added coverage for separate attr_reader/attr_writer declarations