Tune kubespawner to better deal with observed datahub behavior

[yuvipanda]

1. Use curl instead of native tornado for making HTTP requests. This is faster and also makes use of connection pooling.
2. Bump max number of client http connections at a time from 10 to 64.
3. Lower the polling interval for checking if the proxy is ok from 5min to 5s
4. Lower the timeout for a server starting from 20min to 1min. So when spawn fails, users can launch again in 1min and not wait 20min