[derive] Test with trivial_bounds feature

[joshlf]

DO NOT MERGE: We need to debug this issue.

The nightly trivial_bounds causes some derive-emitted code to compile which is designed to fail compilation. This commit adds a test that ensures that the emitted code still triggers some compiler errors.

Affects #61

Closes #500

[stegaBOB]

Hey @joshlf. We rely on static assertions in some of our crates and were considering enabling trivial_bounds as well and ended up stumbling upon this through the RFC issue. In our testing, we couldn't actually get invalid assert_impl_alls to compile. I tried with a few separate Rust versions, and from what I can tell it looks like the order the errors were emitted in the nightly this repo happened to use was unexpected. After satisfying the FromZeros requirement, the assert_impl_all error gets displayed:

error[E0277]: the trait bound `NotZerocopy: FromBytes` is not satisfied
  --> zerocopy-derive/tests/trivial_bounds.rs:60:18
   |
60 | assert_impl_all!(FromBytes1: FromBytes);
   |                  ^^^^^^^^^^ the trait `FromBytes` is not implemented for `NotZerocopy`
   |
   = help: the following other types implement trait `FromBytes`:
             isize
             i8
             i16
             i32
             i64
             i128
             usize
             u8
           and 38 others
note: required for `FromBytes1` to implement `FromBytes`
  --> zerocopy-derive/tests/trivial_bounds.rs:55:10
   |
55 | #[derive(FromBytes, FromZeroes)]
   |          ^^^^^^^^^ unsatisfied trait bound introduced in this `derive` macro
note: required by a bound in `_::{closure#0}::assert_impl_all`
  --> zerocopy-derive/tests/trivial_bounds.rs:60:1
   |
60 | assert_impl_all!(FromBytes1: FromBytes);
   | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ required by this bound in `assert_impl_all`
   = note: this error originates in the derive macro `FromBytes` which comes from the expansion of the macro `assert_impl_all` (in Nightly builds, run with -Z macro-backtrace for more info)

Amusingly, in the nightly 5 days later (2023-10-26 instead of 2023-10-21), all the errors are shown as expected. Presumably some change fixed the error ordering or something.

Is the zerocopy crate still interested in adding trivial_bounds?

[joshlf]

Huh fascinating! I'll have to resurrect this PR and try again.

Yes, we're still interested in trivial_bounds: specifically to support #1769.