Summary

A top-level beforeAll/afterAll/beforeEach/afterEach call creates a synthetic "global" TestSuite to host the hook. Previously that suite was also eagerly registered as a top-level Deno.test, so any nested describe was reported as a child step of global — inflating the step count by one and changing the test output.

For example, this file:

import { assert } from "@std/assert";
import { beforeAll, describe, test } from "@std/testing/bdd";

beforeAll(() => {});

describe("the describe", () => {
  test("test 1", () => assert(Math.random() < 0.0001));
  test("test 2", () => assert(Math.random() === 0.123456));
});

reported FAILED | 0 passed | 1 failed (3 steps) even though only two tests existed. With this PR it reports (2 steps), matching the file's contents.

How the fix works

• The synthetic "global" suite is now only registered with Deno.test lazily — only if a top-level it() is actually added to it. Otherwise the wrapper never surfaces as a counted test/step.
• Each child describe of the synthetic suite is promoted to its own top-level Deno.test. It keeps a back-reference to the synthetic suite (syntheticParent) so the global hooks still wrap its tests at run time:
  • beforeAll / afterAll are invoked around the promoted child's body.
  • The synthetic suite's symbol is pushed onto TestSuiteInternal.active for the duration of the run, so runTest's normal traversal picks up beforeEach / afterEach from the global suite just like any other parent.
• Top-level it() calls under a synthetic suite still work exactly as before: the synthetic suite is registered lazily on the first it() and the it()s become its steps.

Behavior preserved

• Hook-only top-level + nested describe → fixed: nested describes register as their own Deno.tests, hooks still run around their tests (regression test added).
• Hook + top-level it() → unchanged: a single "global" Deno.test wraps the it()s (existing tests still pass).
• it.only() propagation to the synthetic global → unchanged (existing test still passes).
• Mixed (hook + top-level it() + nested describe) → describes are promoted; the synthetic wrapper remains for the top-level it()s. Hooks fire around both, which means beforeAll runs once per top-level Deno.test rather than strictly once across the file. Acceptance criteria only requires the hook still "runs around tests" in this case.

Tests

Added three regression tests in testing/bdd_test.ts:

1. Top-level hooks + only a nested describe → one Deno.test named after the user's describe, two steps, hooks fire correct number of times.
2. Same, but the nested describe also has its own hooks → verifies the global/local hook ordering around each test.
3. Top-level hook + multiple nested describes → each describe is its own Deno.test.

Closes bartlomieju/orchid-inbox#68

Test plan

• Reproduction from the source issue prints (2 steps) instead of (3 steps).
• deno test -A --parallel --no-check testing/ — 137 passed (159 steps), 0 failed.
• deno test -A --parallel --no-check expect/ — 167 passed, 0 failed (expect builds on bdd).
• deno fmt --check — clean.
• deno lint — clean.
• deno test --doc testing/bdd.ts — doc examples still pass.