Fix rarely crash when try to call c4socket_gotHTTPResponse on sock_id…

[CedricCouton]

… with connection already closed

Try to fix #54

[Dushistov]

Sorry for delay, I was extremely busy with another projects.

If I get you right, you call restart_replicator with very short delay,
so close initiated on the client, not on the server,
so we get situation:

1. ws_open, which return control back, but run async work to finish initialization
2. ws_request_close which close socket before async work that start ws_open would have chance to be finished.

I got it right?

Then such construction should only mask problem, not solve it:
if !*socket.closed.lock().await {

I mean mutex at this line locked and unlocked, all in the same line of code. So if close happens between lines 365 and 379 this would be bad?

May be better have some kind of barrier in async part of ws_open, so until we reach 'read_loop: loop {
async part of ws_request_close will wait. May be in busy loop, like:

init_done: Arc<AtomicBool>,

open_connection:
   init_done.store(true);

ws_requse_close:
    while !init_done.load() {
     }