Persistent connections for reference fetching

[martinthomson]

I just had a build fail due to a DNS error after successfully fetching resources from the very same server.

/github/home/.cache/xml2rfc/reference.RFC.7838.xml: fetching from https://www.rfc-editor.org/refs/bibxml/reference.RFC.7838.xml
/github/home/.cache/xml2rfc/reference.RFC.6265.xml: fetching from https://www.rfc-editor.org/refs/bibxml/reference.RFC.6265.xml
*** Failed to open TCP connection to www.rfc-editor.org:443 (getaddrinfo: Try again) while fetching https://www.rfc-editor.org/refs/bibxml/reference.RFC.6265.xml

This would not happen if the HTTP client were able to reuse connections. I realize that this is work, but it would make the tool considerably faster.