Use connection pooling for Python client's requests

I was using cProfiler with the Python client to help diagnose PLFM-4836 and discovered that the python client was spending about 1-2 seconds per request performing an SSL handshake.
While talking with , he suggested that I configure the requests library to use connection pooling when contacting the Synapse backend. This change would allow the Python client to reuse the same connection for successive requests and reduce the amount if time it takes for each API calls to complete after the SSL handshake is completed.