regression: null bytes in utf8-stream

[Mark-Simulacrum]

[INFO] [stdout] thread 'test_iter_new' (18) panicked at tests/test_utf8_stream.rs:8:5:
[INFO] [stdout] assertion `left == right` failed
[INFO] [stdout]   left: ["\0", "\0", "\0", "\0"]
[INFO] [stdout]  right: ["t", "e", "s", "t"]

https://crater-reports.s3.amazonaws.com/beta-1.92-3/beta-2025-11-01/reg/utf8-stream-0.0.0/log.txt

There's a bunch of tests from this crate where ASCII text is now null bytes in the test assertions. The crate has a bunch of unsafe code so it's plausible this is just benign UB exposing itself, but we should confirm that it's not a breaking change (e.g., bisect or run under Miri).

[saethlin]

Both Miri and ASan indicate that utf8-stream is doing use-after-free. Bisection points to #142771, which is a bit odd but it does touch a lot of codegen.

I don't think it's fair to call this "benign" UB because if you set -Copt-level=1 the tests fail on stable 😉