EII: only require implementations for final artifacts to support -C prefer-dynamic

[cezarbbb]

EII checking currently uses CheckExistence for any crate type that isn't rlib, which means dylib/sdylib/proc-macro crates that declare an EII without an implementation always fail to compile. But these are intermediate products — their EIIs can be resolved by downstream crates, just like rlibs. The linker selects the strong symbol (explicit impl) over the weak one (default impl) at final link time, so intermediate crates don't need to guarantee an implementation exists yet.

This changes get_checking_mode() in rustc_passes/src/eii.rs to only use CheckExistence when the crate produces a final artifact: Executable, Cdylib, or StaticLib. All other crate types (Rlib, Dylib, Sdylib, ProcMacro) only check for duplicate explicit implementations. This matches the existing CrateType::has_metadata() classification in rustc_hir/src/attrs/data_structures.rs — crate types with metadata are intermediate products that downstream crates can depend on, and those without metadata are leaf nodes in the dependency graph where no further resolution is possible.

With this change the behavior for each crate type is:

• Rlib / Dylib / Sdylib / ProcMacro — no explicit impl is fine, compiles successfully; duplicate explicit impls across dependencies still error
• Executable / Cdylib / StaticLib — every EII must have at least one implementation (default or explicit), otherwise EiiWithoutImpl error; duplicate explicit impls still error

This modification's result is the -C prefer-dynamic flag works: a dylib intermediate can declare an EII without an implementation, and the final executable provides the explicit impl — the strong alias overrides the weak one at link time.

tracking issues: #125418

r? @jdonszelmann

[rustbot]

jdonszelmann is currently at their maximum review capacity.
They may take a while to respond.

[bjorn3]

This effectively reverts #156319. I don't think it is correct, see the zulip thread: https://rust-lang.zulipchat.com/#narrow/channel/213817-t-lang/topic/EII.3A.20should.20dylib.2Fsdylib.2Fproc-macro.20require.20implementations.3F/near/606144824

[cezarbbb]

I agree with bjorn's reply in zulip. I will close this PR and explore new ways to enable EII support prefer-dynamic.