Add sections for wire image, protocol choice, and cost

[tfpauly]

This PR attempts to tackle several issues, in related areas of the text:

• Explain cases where greasing can be negotiated, for the benefit of modifying the wire image (QUIC bit case) (Discuss greasing the QUIC Bit (as an unencrypted field) #5) (@dthaler's point in Highlight that greasing applies when received unknown values are ignored -- applicability! #50)
• Describe aspects of wire image variation that are independent of specific protocol fields (Wire image variation #53)
• Variation in choosing which protocol to use, or whether or not to use a protocol (Greasing considerations to endpoint features #3)
• Describe costs of variation on efficiency (Wire image variation #53)

[LPardue]

I think I understand your intent in this specific part text but the nuance might be lost.

The prescence (or absences) of observers on the visible wire image doesn't always affect the choices endpoints might make wrt exercising variation. The QUIC spin bit is maybe an example of that. Endpoints don't negotiate whether its set or not, but twiddling the bit is a way to exercise greasing onto the observers.

The QUIC bit is an interesting example, because it serves a very specific purpose (demultiplexing QUIC in deployments that need to). The negotiation is therefore needed to disable that bit, without affecting the QUIC protocol, in deployments it is known that the demultiplexing is not needed. The QUIC bit may be the wrong thing to reference here to support the point.

[tfpauly]

The point I'm trying to make here is that:

Some fields require negotiation to change (like the QUIC bit). That doesn't help grease the peer at all. However, for fields that are visible to the wire image, the greasing is still useful. That's the only reason that greasing the QUIC bit makes sense.

[tfpauly]

@LPardue please check out e6365bf

[LPardue]

Thanks, that reads much clearer to me!