Fix websocket reconnect race condition

[yschimke]

Attempting a fix for #7381

The reason why it happens when fast reconnecting is that when we fail because of a 401, we release the Connection back to the ConnectionPool. The new connection started from the onFailure callback is very likely to get the same connection, and so two threads are now operating on the same socket.

    call!!.enqueue(object : Callback {
      override fun onResponse(call: Call, response: Response) {
        val exchange = response.exchange
        val streams: Streams
        try {
          checkUpgradeSuccess(response, exchange)
          streams = exchange!!.newWebSocketStreams()
        } catch (e: IOException) {
          exchange?.webSocketUpgradeFailed() // releases the connection back to the ConnectionPool
          failWebSocket(e, response)
          response.closeQuietly() // Drains the Response data
          return
        }

[yschimke]

@swankjesse requesting post review, would like to test with a published snapshot.

From slack "fix looks good"

[yschimke]

Damn, I took a success as a pass but it was really a lucky run of a flaky test

[swankjesse]

LGTM