Resolve the false positive dead_code lint when type is only used via destructuring

[m-ysk]

Fixes #132874

Resolves the false positive dead_code lint when type is only used via destructuring reported by the issue #132874 .

The existing code seems to have forgotten to mark the variant itself as a live symbol while it marks the variants fields. This PR adds the code to mark the variant itself as a live symbol in order to resolve the false positive dead code.

[rustbot]

r? @nnethercote

rustbot has assigned @nnethercote.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[m-ysk]

@nnethercote
Hello. I've done all the works I can do for now though I'm not so confident. Could you review this?

[nnethercote]

Hmm, I'm not sure about this change. Because the old statement that Y is never constructed is correct. And if a variant appears in a pattern but is never constructed I think it should be warned about, because that could genuinely indicate dead code.

So now I'm questioning the whole premise of #132874 and this PR. If anything, I think let Foo(_x) = get_thing::<Foo>(); should be made to give a warning. Does that seem reasonable?

[nnethercote]

Also, it would be better if the commits were squashed together because they are all part of the same logical change. (But that may be moot if the PR doesn't go ahead.)

[nnethercote]

(Also, you've done a nice job on the code and test changes, they look good if we are willing to go in this direction.)

[m-ysk]

Thank you for your review. I understood that the code given in the issue #132874 should be warned as you explained.

Now, I'm thinking that there's a possibility that the issue owner actually found a bug, but the MRE in the issue was over-generalized because he said he found the issue when using zerocopy::FromBytes::read_from_bytes and that function contains extra language features (derive macro, trait) that have been missed in the MRE.
So, I'm asking to give us another MRE closer to the original code.
#132874 (comment)

[m-ysk]

@nnethercote
The issue author gave us another repro as follows, which is closer to the real code.
#132874 (comment)

The new code indicates there are cases where a struct is actually used without being constructed in the normal way. I think this code should be compiled without any warning.

Though the best is to achieve both allowing the above code to be compiled without any warning and keeping the warning to the cases like tests/ui/lint/dead-code/lint-dead-code-4.rs, I'm not sure whether it can be achieved.

So, can I have your advice on which direction to go?

[nnethercote]

That is a tricky case: a PacketHeader is constructed, but via binary deserialization, which the compiler cannot understand. It seems like a false positive that is hard to avoid without introducing false negatives in other cases. I'm not sure how best to handle it. I have asked for other opinions in this Zulip thread.