testing: effectTest harness — return an Effect, get pass/fail

[GraemeF]

What

Adds effectTest to @commy/testing — a bun:test harness that lets a test body return a success/failure Effect instead of hand-wrapping Effect.runPromise(Effect.scoped(Effect.gen(...))) at every call site.

Foundation for the hexagonal test architecture (comms-e5vm); first consumer is the Tier-2 proof (comms-e5vm.2).

Why

Nearly every test in the suite repeats the Effect.runPromise(Effect.scoped(...)) wrapper, and a failing Effect rejects with a FiberFailure that buries the real error/assertion behind the promise boundary — so a failed expect() shows up wrapped, not as itself.

How

effectTest(name, body, options?):

• always wraps the body in Effect.scoped, giving every test a per-test Scope whose finalizers run on both the success and failure paths;
• options.layer threads the test's requirements (R) without a global — the caller composes captureLogger / TestContext (TestClock) / the stub HttpClient layer (comms-e5vm.1) and passes them as one layer. The base harness stays on the live clock, so TestClock is opt-in and live-clock tests are unaffected;
• runs via Effect.runPromiseExit and, on failure, throws Cause.squash(cause) so the raw error surfaces. squash (not prettyErrors) deliberately returns the original thrown object untouched, keeping bun's assertion-diff renderer intact.

runTestEffect is the engine behind the registrar, exported so the pass/fail mapping is itself testable (and usable inside a custom test.each).

Design decisions (resolving the bead's open questions)

• One function, no .scoped / .live variants — the base is already always-scoped + live-clock; everything else is opt-in via options.layer.
• Generic layer slot, no build-time dep on comms-e5vm.1 — the harness never knows the stub's concrete shape; the two compose at the consumer's call sites. Keeps both foundation beads landing in parallel.
• Migration is out of scope (adopt-going-forward) — that's comms-e5vm.2.

Tests

11 tests covering: success resolves with value; typed failure and thrown defect both surface the raw object (not a FiberFailure); per-test Scope finalizers run on success and failure; layer provision via a Context.Tag service, captureLogger, and TestContext/TestClock; layer-build failure surfaces; plus the effectTest registrar wired end-to-end.