inference: make `Core.Compiler.return_type` respect `max_methods` setting

[aviatesk]

Previously we didn't limit the number of matching methods for Core.Compiler.return_type, which could lead to a potential latency problem. This commit makes Core.Compiler.return_type and the corresponding tfunc return_type_tfunc respect max_methods setting as like the ordinary inference. One caveat here is that the current Core.Compiler.return_type interface unfortunately doesn't allow it to see the caller module context, so it can't respect module-wide max_methods setting (I guess the change to the interface should be made as of version 2.0), that I think can lead to a potential confusion.