chore: Converted agent unit tests to node:test

[jsumners-nr]

This PR resolved #2380.

There are couple of main differences between tap and node:test patterns:

1. In node:test every test() is actually a promise.
2. In node:test every test function is wrapped in a promise.
3. In node:test there is not a subtest .end or .autoEnd concept.

So if we have a tap test like:

tap.test('some stuff', t => {
	t.test('one', t => {
		t.equal(1, 1)
	})

	t.test('two', t => {
		foo('bar', (error, val) => {
			t.error(error, 'no error should occur')
			t.equal(val, 'bar')
			t.end()
		})

		function foo(param, cb) {
			cb(param)
		}
	})
})

The equivalent in node:test would be:

test('some stuff', async t => {
	await t.test('one', t => {
		assert.equal(1, 1)
	})

	await t.test('two', async t => {
		const { promise, resolve } = Promise.withResolvers()

		foo('bar', (error, val) => {
			assert.equal(error, undefined, 'no error should occur')
			assert.equal(val, 'bar')
			resolve()
		})
		await promise

		function foo(param, cb) {
			cb(param)
		}
	})
})

[jsumners-nr]

This was devised to replace the usage of nock in test/unit/agent/agent.test.js. I encountered an issue where one of the subtests was unable to be resolved due to a timing issue between promise resolution and the event loop finishing. Utilizing "real" network requests helped me resolve the problem. If I remember correctly, it was this test causing grief:

node-newrelic/test/unit/agent/agent.test.js

Lines 494 to 504 in 09636a4

	t.test('should not blow up when harvest cycle runs', (t) => {
	agent.start(() => {
	setTimeout(() => {
	t.ok(redirect.isDone())
	t.ok(connect.isDone())
	t.ok(settings.isDone())
	t.end()
	}, 15)
	})
	})

Maybe the test could have been reworked with nock still being used, but I am less convinced of nock's utility nowadays than I used to be. While the pattern is not used by any tests in this PR, by using the approach in this PR we can do things like:

test('something', async t => {
	const { collector } = t.ctx
	const { promise, resolve = Promise.withResolvers()

	collector.addHandler('/some/endpoint', (req, res) => {
		// Verify that the client generated the correct request structure:
		assert.equal(req.url, '/some/endpoint?with=query_params')
		res.writeHead(200)
		res.end()
		resolve()
	})

	await promise
})

Yes, it is possible with nock, but, despite using it for many years, I find myself having to go reference the documentation every time I want to do something like that.

[jsumners-nr]

This should be generic enough to utilize in all tests that need to talk to the remote collector. I haven't actually looked, but I bet we can replace the fake-collector references in test/integration/collector-remote-method.tap.js with this.

[jsumners-nr]

This sinon.restore() is the sole reason for updating to sinon@5 (and evidently sinon@18 is the current version). Basically, in sinon@5 the default import is itself a "sandbox" and thus obviates the need to const sandbox = sinon.sandbox() and do a dance around that for this simple test.