Simplify SETTINGS payload

[MikeBishop]

Fixes #1556 by making all setting values varints and removing length prefixes. This has two ramifications, which are probably livable:

• An extension which needs a structured configuration message needs to define an extra frame/stream type to carry that message.
• An extension which used the full 32-bit space in HTTP/2 will spill over to a 64-bit varint, because in four bytes we only carry 30 bits of payload.

[ianswett]

I'm not sure I understand the purpose of this additional text? You lose 2 bits on 32 bit fields, but do we really expect that will require changes on anyone's part?

[MikeBishop]

Secondary Certs currently stuffs an exporter in the setting value (to confirm both peers are on the same TLS connection with no MitM), and would overflow to an eight-byte field. It doesn't require changes, but an HQ port would either take advantage of the entire eight bytes or trim two bits off to still fit in four bytes. Hence, "might choose."

[ianswett]

In that case, how about starting the paragraph with 'Setting values are ...... rather than fixed length 32-bit fields as in HTTP/2." Currently you mention HTTP/QUIC in the HTTP/QUIC draft, and don't reference the thing you're comparing against(HTTP/2).