feat(runner): add afterAll hook for describe.serial block cleanup

[monikon22]

Problem

describe.serial has no hook that's guaranteed to run once the block is done, whatever "done" turned out to mean. afterEach/ctx.cleanup() are per-test: once the block stops early (a failure, a timeout, invalidatePlayer), every remaining test is reported skipped and its body — and any cleanup registered inside it — never runs. There's no way to write "revoke this permission / release this resource no matter which test in the block failed."

Fix

Adds afterAll(hook), importable alongside test/describe/beforeEach/afterEach. Only valid inside describe.serial — a plain describe/test has no single block-completion event to hang it on, and calling it outside a block throws immediately at registration time.

It runs:

• once per instance of the block (so concurrency: 5 runs it 5 times, once per bot)
• after the last test that actually executed — whether the block finished cleanly, stopped early, or a test in it was skipped
• before the plugins' own afterEach and before the bot disconnects
• not at all if the block never got its player (the initial connect failed) — nothing ran, so there's nothing to clean up

A throwing afterAll hook is logged ([serial <block>] afterAll error: ...) and never flips the block's result, matching how ctx.cleanup() finalizer errors are already handled.

describe.serial('vip trial', () => {
  test('grants the trial permission', async ({ player, server }) => {
    await server.execute(`lp user ${player.username} permission settemp vip.trial true 15m`);
  });

  afterAll(async ({ player, server }) => {
    // Runs once, even if a later test in the block fails or gets skipped.
    await server.execute(`lp user ${player.username} permission unsettemp vip.trial`);
  });

  test('sees the vip shop', async ({ player }) => { /* ... */ });
});

Changes

• test-registry.ts: SerialBlock.afterAllHooks, afterAll() export
• test-runner.ts: runSerialBlock runs the block's afterAllHooks (LIFO, try/catch-logged) in its finally, gated on lastCtx (mirrors the existing plugins.afterEach(lastCtx) gate)
• runner.ts: re-exports afterAll
• docs/writing-tests.mdx, docs/api-reference.mdx: document the hook

Verification

tsc --noEmit clean. No existing behavior touched outside the runSerialBlock finally block and the registry's block construction.

[Drownek]

Nice addition, this fixes a real architectural pain point in describe.serial test suites.

What I checked:

• Guaranteed cleanup: this fixes the issue where an early test exit or failure would skip downstream test bodies, leaving server-side state (permissions, whitelists, economy data) dangling with no cleanup.
• Lifecycle & concurrency: hook execution in the finally block, scoped per bot instance with its own lastCtx, holds up fine under concurrency: N.
• LIFO ordering: reversing the hook array gets teardown order right, matching how resource unwinding normally works.
• Resilience: catching and logging throwing hooks stops a broken teardown from flipping an otherwise-passing test block.
• Graceful skip: correctly skips execution when the bot never connected (lastCtx === null).
• Registration guard: calling afterAll outside describe.serial throws immediately, so you find out at import time instead of during a run.
• Docs in api-reference.mdx and writing-tests.mdx are clear.

Tested locally with early test exits, behavior checks out.

LGTM, ready to merge.