Switch to amphp/amp v3

[przepompownia]

No description provided.

[przepompownia]

Please keep in mind that I'm still learning cooperative multitasking/AMP{2,3}/Revolt/Fibers and still finding basic examples to dispel numerous doubts (especially related to transition from v2 to v3).

I started rehearsing with the phpactor/language-server project, then stashed the changes and switched here since it's probably the simplest dependency placed between Phpactor and AMP.

Not sure if it's a good approach, but (instead of tests) I started from the demo executable (./bin/watcher) and InotifyWatcher, then after (perhaps apparent) success updated FindWatcher.

Expect elementary mistakes. Here I still need a hard proof (a test) that all previously non-blocking operations remain so. Early feedback is also welcome, so I've released this change as a draft.

[przepompownia]

Still unsure if (on watchers) we need to wrap wait body with async and await its Future somewhere outside - unfinished loop on Phpactor\AmpFsWatcher\Tests\Watcher\Find\FindWatcherTest::testSingleFileChange (and other cases) causes exceeding timeout 5s when running on my local machine.

An alternative is an external wrap - for example at the test (each case) level

    public function testSingleFileChange(): void
    {
        $process = $this->startProcess();
        $this->delay();
        $this->workspace()->put('foobar', '');
        $this->delay();

        async(function () use ($process): void {
            self::assertEquals(
                new ModifiedFile(
                    $this->workspace()->path('foobar'),
                    ModifiedFile::TYPE_FILE
                ),
                $process->wait()
            );
        });

        $process->stop();
    }

and (maybe accidentally) allows to satisfy the assertion and stop the process.

[przepompownia]

Both in the bin/watch demo and test cases I turned wait() into async outside its body.

For some reason that is still unclear, the test stopped displaying the unfinished loop message after increasing the delay between operations.

@dantleech what do you think about the current state?