Python's httplib generally has bad support for Keep-Alive connections,
which is why we never used it. We didn't explicitly ask for "Connection: Close",
but it was implicit. Additionally, splunkd would always return it.

However, an issue arises when talking to a Load Balancer which issues
a Keep-Alive request to splunkd, and returns that header to us. Python
then goes into a particular path which the code didn't handle gracefully.

What this change does is two-fold:

1. Always request to close the connection via the "Connection: Close" header.

2. If we get back a "Connection: Keep-Alive" header, we properly handle
it by not immediately closing our reference to the connection and only
doing so when the response has been completed.