Fix current randomly failing unit tests

[pathscale]

Some of tests are failing sometimes (not on every run) and backtrace leads to indexset dependency or UnsizedNode object. We need to fix this issues as this can lead to bugs in runtime

---

Migrated from pathscale/wt-dead#129. Originally opened by @Handy-caT on 2026-01-08T20:57:55Z.

[pathscale]

Reproduced on clean origin/master. One integration run failed after three passes; a captured stress sequence then failed on run 19 after 18 passes. Exact failure: worktable::unsized_::update_parallel_more_strings_with_select_unique panicked in WorkTablesIndex 0.0.4 concurrent/set.rs:524 with target was found while the node was locked. The audit also completed 42 consecutive passing runs between failures, confirming nondeterminism. This remains a 1.0 blocker until the WTI invariant violation is fixed and the WorkTable pin is updated, followed by a high-count clean loop.

[pathscale]

The original reproducer now looks resolved by WorkTablesIndex 0.0.5 as consumed by PR #46. worktable::unsized_::update_parallel_more_strings_with_select_unique passed 100/100 fresh-process repetitions on the PR merge ref, and the same test passed current CI. This satisfies the requested high-count clean loop for the former WTI locked-node assertion. Recommend closing when PR #46 lands.

[pathscale]

Resolved by WorkTablesIndex 0.0.5 as consumed by merged PR #46 and worktable 1.0.0-beta.3. The original unsized unique-select/update reproducer passed 100/100 fresh-process repetitions, and the final default, versioned-publication, and all-features CI jobs passed.