fix: Hermes withPrefixKey crash and Turbopack static this in scope.module

[ssalbdivad]

Closes #1645
Closes #1643

• Disjoint.withPrefixKey now rebuilds a Disjoint when map returns a plain Array, the same Hermes workaround invert already had (Unable to use ArkType with Expo #1027). A test simulates Hermes by overriding Symbol.species.
• InternalScope.module calls InternalScope.scope instead of this.scope, which Turbopack can compile to an undefined binding.

Lint, build and testRepoWithVersionsAndBenches pass locally.

🤖 Generated with Claude Code

[pullfrog]

✅ No new issues found.

Reviewed changes — verified both Hermes/Turbopack compatibility fixes and the accompanying regression test.

• Disjoint.withPrefixKey Hermes workaround — rebuilds a Disjoint via new Disjoint(...result) when map returns a plain Array, mirroring the existing invert guard. The removed as Disjoint cast is correctly replaced by a truthful as DisjointEntry[]; the instanceof check then narrows back to Disjoint. On standard engines the behavior is unchanged.
• InternalScope.module static this removal — this.scope is now InternalScope.scope. The old this resolved identically for every live caller (InternalScope.module, the re-exported Scope alias in ark/type/keywords/*, and the detached $.constructor.module attached to Parser), so this is behavior-preserving while removing the Turbopack static-binding hazard.
• Regression test — overrides Symbol.species to Array and asserts both withPrefixKey(...) and invert() return Disjoint instances. I reverted the withPrefixKey guard locally and confirmed the test fails (false !== true at union.test.ts:112), so it pins the actual bug rather than passing vacuously.

Other Disjoint methods that use map (describeReasons) only call .join, so they don't depend on subclass preservation and need no equivalent guard — coverage here is complete.

  ^{｜ View workflow run ｜ Using deepseek-v4.1-flash (free via Pullfrog for OSS) ｜ 𝕏}