Don't require 1:1 mapping between OpTypeStructs and their names.

[eddyb]

The linker deduplicates types based on their definition anyway (though maybe we should do less of that - see #398 (comment)), so the main side-effect of SpirvType::Adt having contained a name: String was that some types would be considered different during codegen itself (even when otherwise compatible). Specifically, the added test used to fail with:

error: Cannot cast between pointer types
   --> /home/eddy/.cargo/registry/src/github.jparrowsec.cn-1ecc6299db9ec823/glam-0.12.0/src/mat3.rs:462:1
    |
462 | impl_mat3_traits!(f32, mat3, Mat3, Vec3);
    | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    |
    = note: from: *{Function} struct spirv_std::glam::Mat3 { 0: struct spirv_std::glam::core::storage::Vector3x3<spirv_std::glam::XYZ<f32>> { x_axis: f32x3, y_axis: f32x3, z_axis: f32x3 } }
    = note: to: *{Function} struct spirv_std::glam::core::storage::Vector3x3<spirv_std::glam::Vec3> { x_axis: f32x3, y_axis: f32x3, z_axis: f32x3 }

But with this PR, a single type definition is generated, and it gets two OpNames, instead:

OpName %5 "spirv_std::glam::core::storage::Vector3x3<spirv_std::glam::XYZ<f32>>"
OpName %5 "spirv_std::glam::core::storage::Vector3x3<spirv_std::glam::Vec3>"

(snippet obtained via DUMP_PRE_LINK, to avoid linker deduplication, #398, etc.)

If we had some compile failure tests akin to ui tests in rust-lang/rust, it would've been useful to see how bad the "show the first name" approach can get. If we think printing SpirvTypes in errors is relatively rare, we could show all names even etc.