fix(cli): synchronize async stdio/file reads and writes

[dsherret]

Fixes a regression probably caused in #14319 where asynchronous file reads and writes no longer necessarily wrote in order.

Closes #14353

[dsherret]

Seems like this is still flaky. I can't reproduce it as flaky on my system on Windows or Linux though https://github.com/denoland/deno/runs/7205374557?check_suite_focus=true

Edit: Oh, I reproduced it on linux after about 10 minutes of re-running the tests over and over.

[dsherret]

Just verified that this is called while still executing synchronously from the JS write(...) call. Also, on the line immediately before the acquire here, the data is in order:

Hello, world! 0
Hello, world! 1
...
Hello, world! 20
Hello, world! 21
Hello, world! 22
...
Hello, world! 98
Hello, world! 99

But then after the semaphore it's (I'm not sure what xa is):

Hello, world! 0
Hello, world! 1
...
Hello, world! 20
Hello, Hello world!
Hello, world! 22
...
Hello, world! 98
Hello, world! 99
world! 21xa

Also, this always occurs at 21.

[lucacasonato]

Can TaskQueue not be a runtime/ thing? I don't think it needs to be in deno_core.

[dsherret]

@lucacasonato thanks for the offline feedback! I managed to switch back to an AsyncRefCell and eliminate all the mutexes being used here. I also reduced some locking on std::io::... stuff.

[lucacasonato]

Looks great now. If you can easially add a test for the "sync read while async read is ongoing" case, that'd be great. If it's too difficult, feel free to merge without that.