fix(js): keep foreign effects and scheduling helpers together

[tachytelicdetonation]

Fixes #946.

An effect named A.run_need can overwrite A.run's scheduling helper in the flat JS foreign table, or be overwritten by it. Store one {run, need} record per effect so both functions retain their module's identity regardless of discovery order.

Use computed property keys so __proto__ is also stored as an own effect entry, covering the additional reproduction reported by @jasisz on the issue. Existing module isolation and duplicate effect-name checks remain in place.

The compiler change is 4 lines added / 5 removed (+12 ttok). Three regression cases cover both discovery orders and __proto__; the scheduling fixture checks that its helper was invoked without relying on elapsed-time assertions.

Verification:

• Both collision orders failed before the fix; __proto__ also failed before the computed-key change. All three now pass through the default runner and emitted JS.
• 13 focused tests pass on macOS arm64 and Linux arm64 (34 checks per platform), including shared module state, runtime-name isolation, foreign arity/types, timers, and parked TCP/UDP effects. Native C checks run where fixtures provide a C implementation.
• All 16 runtime benchmarks emit identical C; their emitted JS differs only in the intended foreign-table/request wiring. Repository file/token gate: 46/46; git diff --check passes. TypeScript diagnostics match the baseline's four existing errors.

The private mini-cluster test and performance gates were not available. JS checks used Bun; no cluster timing results are claimed.

[VictorTaelin]

Merged into 2.0.25 under your authorship, thank you. Both discovery orders and the __proto__ effect now print the right values on all lanes; we audited every reader of the foreign table against the {run, need} record. This closes #946.

Note: this reply was written by an AI after it reported the issue to me and I made the decision. If anything here is wrong, reply and I will review it myself.