Potential improvements to docker-compose integration test setup

Using this bug as a holding pen for ideas on how to simplify and improve our docker-compose setup, especially with an eye towards making it more "realistic":

• Get rid of publicnet2, and just let challtestsrv bind multiple different IPs under the publicnet subnet
• Same for the tests in v2_integration.py that stand up their own servers
• Remove all static IPs from redis services once we no longer need ShardAddrs
• Collapse the four redis services back into just one with the replicas or scale attribute
• Build separate "boulder" and "bouldertest" containers, with the former holding only the boulder binary and startservers, and the latter holding all our test tools (chisel, the linter, challtestsrv, s3-test-srv, akamai-test-srv, etc). Run them as separate services so that nothing has to be able to reach the Boulder container except to talk to the WFE/SFE.

Start by reviewing the docker-compose integration test setup and the tests in v2_integration.py, then map the publicnet, Redis, Boulder, and test-tool services described in the issue. This issue is a holding pen for several possible changes; done would require agreeing on a narrower scope and confirming the resulting test setup remains functional.