Add an option to box oneof fields of kind message

[hurryabit]

Currently, protoc-rust cannot handle mutually recursive messages with oneofs, even with singular_field_option_box enabled. For instance, the example

message A {
    oneof a {
        int32 i = 1;
        B b = 2;
    }
}

message B {
    oneof b {
        int32 i = 1;
        A a = 2;
    }
}

produces Rust code whose compilation fails with

error[E0072]: recursive type `common::v2::test_oneof_recursive_pb::A` has infinite size
   --> protobuf-test/src/common/v2/test_oneof_recursive_pb.rs:273:1
    |
273 | pub struct A {
    | ^^^^^^^^^^^^ recursive type has infinite size
274 |     // message oneof groups
275 |     pub a: ::std::option::Option<a::A>,
    |     ---------------------------------- recursive without indirection
    |
    = help: insert indirection (e.g., a `Box`, `Rc`, or `&`) at some point to make `common::v2::test_oneof_recursive_pb::A` representable

For recursive messages that do not go through other messages in the recursive cycle, this problem has been fixed by an ad-hoc check to detect these messages and adding Boxes where necessary:

rust-protobuf/protobuf-codegen/src/oneof.rs

Line 35 in 0674fbb

// detecting recursion

Unfortunately, this recursion detection does not easily extend to mutually recursive messages.

This PR proposes to solve the problem by allowing the user to specify when oneof fields should be boxed. Therefore, we introduce an option oneof_field_box that works like singular_field_option_box but on oneof fields of kind message. Fortunately, the infrastructure for doing the boxing/unboxing has already been implemented for the existing ad-hoc recursion check. Thus, we only need to pass the option into the right place and add a few tests.

---

This change is 

[stepancheg]

Having an option is OK, but I think this case should work automatically without additional options specified.

Unfortunately, this recursion detection does not easily extend to mutually recursive messages.

I think it's doable.

I need to try to implement it, and if you want to try it yourself, go ahead.

[stepancheg]

I have implemented recursion detection in 5d70656.

Is oneof_field_box still necessary?

Keeping this PR open because this commit needs to be backported to the stable branch.

Thank you for the bugreport/PR!

[stepancheg]

Ported to 2.7 branch c4ca2e9.

2.7 will be released in ~1-2 weeks. Please let me know if this is critical and need to be released earlier.

[stepancheg]

I'd apprecate code review BTW, because there's a chance I overlooked something.

[hurryabit]

@stepancheg I had a look at your fix. The code looks reasonable and simpler than I expected. Thanks a lot for fixing this so quickly!

I'm currently using a pinned version from the master branch. A quick release of 2.7 is hence not critical for me.