test(core): narrow effect flock harness - #41837

What

Cuts the EffectFlock suite's measured median runtime from 6.45 seconds to 3.64 seconds, about 44%, by avoiding an unrelated Core Location import graph in every lock worker.

Before / After

Before: The EffectFlock test and worker used AppNodeBuilder, whose eager import of Core Location services loaded the complete application graph into the parent process and each of 16 contention workers. EffectFlock itself only depends on Global and FSUtil.

After: Both boundaries compile EffectFlock.node directly with LayerNode.compile. Dependency provisioning remains explicit and identical, without importing Location services.

How

• Replaces AppNodeBuilder.build with LayerNode.compile in effect-flock.test.ts.
• Applies the same focused construction in effect-flock-worker.ts, where repeated process startup makes the import graph especially expensive.
• Leaves worker count, lock timings, production code, and assertions unchanged.

Scope

This is a test-harness import-boundary change only. It does not reduce contention coverage or alter EffectFlock behavior. Reducing the 16-worker stress count is a separate coverage tradeoff and is intentionally excluded.

Testing

• Audit baseline: median 6.45s, range 5.82-8.58s
• Focused result: 1 warmup + 7 measured runs, median 3.64s, range 3.44-5.41s
• Benchmark: 96 passed across 8 suite runs
• Focused suite: 12 passed
• bun typecheck from packages/core
• Push hook: typechecked all 39 workspace packages
• Oxlint, Prettier, and git diff --check