Organize gossip random packet generation

[alexpyattaev]

Problem

Gossip's random packet generation was never designed to be reproducible, and did not cover all variants of all packets. As a result, it is really hard to make reproducible pseudorandom gossip packets for purposes such as wire format validation and to track down possible incompatibilities between implemenations.

Also all of that code was not guarded by dev-context-only-utils feature-gate, making it accidentally usable in prod code.

Summary of Changes

All of the changes here concern only the code that runs in tests and benches.

• Added testing_fixtures.rs to organize the test-related code better
• Added FormatValidation trait to organize the new_rand functions for later use in wire format example generation
• Extended the new_rand implementations where appropriate to cover all variants of all messages

This intended to be the first step to address #4879

[behzadnouri]

Lets hold on here.
We need to discuss what is the end-goal, and what are the alternatives.
If the end goal is to verify wire formats there are better and more accurate ways to do it.

The new_rand methods were not meant to create any realistic values, and are not suitable for this purpose.
Also not clear what is the purpose of FormatValidation trait and why do we need that, and imposing that on CrdsValue would be a maintenance burden.

[behzadnouri]

To provide transparency to other folks, this was discussed over zoom/slack.
Some notes are added on the github tracking issue: #4879