A number of old, nagging issues have persisted since the early days of this codebase. A coherent collection of them represents a set of conjoined functionality designed primarily to support development, and specifically the development practices of those days, but create undue complexity for the more mature state of today, particularly in the areas of node discovery, treasure map (or other incidental payload) handling, the logic specific to "firstula" situations, and test logic.

This PR is an attempt to address a large enough subset of these issues to represent a "new normal" for the codebase, without doing so much as to leave readers confused.

What follows is a very high concentration of impact-per-line and issue-days-destroyed.

Faster failing for everyone

It is now much more difficult to accidentally proceed into the logic of any Character's method if that logic will eventually fail because no nodes are available. It is no longer necessary to call start_learning_loop to achieve failfast functionality. Fixes #2110.

Vladimir's comeback

Vladimir can now attempt to store a fraudulent TreasureMap on a target Ursula. Since we now restrict the size of uploads, this attack might be imagined as splitting up an HD movie into a few thousand small pieces and scattering them across the network.

In the course of ensuring that this attack is properly demonstrated and mitigated, all of Vladimir's logic has been refreshed and touched up. This fixes #1075.

Bob's favorite nodes

Where do we store an incidental payload (TreasureMap is the simplest example, but we also want forward-compatibility with others), we do so only on a tiny subset of nodes. These nodes are selected on the basis that the second character in Bob's encrypting trinket matches on of the first few characters of the node's checksum (perhaps soon canonical - see #1995). Fixes #342.

Alice signs maps with her wallet keypair; Ursula verifies against HRAC.

The do_store logic for TreasureMaps now includes a check that the map matches a published policy, and that the author of the policy has signed the map. Fixes #429. Fixes #1736.

Tests now use "hardcoded teachers" as seed node logic.

This was previously not possible because many tests (and in fact the fixtures) left the MOCK_KNOWN_URSULAS_CACHE unclean. Now, each test is responsible for this cleanup, and an error is raised by the major fixtures (federated_ursulas and blockchain_ursulas) if the cache is unclean at setup time.

New asynchrony patterns

Alice now enacts a Policy using a NodeEnagementMutex, which performs network operations asynchronously and has two blocking methods - one which returns after five percent completion and another which returns after conclusion.

Smaller changes

• If port selection in a test is above 64000, we try again. ValueError: Port out of range 0-65535 #1546
• Announced nodes are no longer verified in crosstown_traffic. This fixes node_metadata_exchange basically reinvents parts of remember_node and learn_from_teacher_node #555 and the ursula_in_a_house problem - though it was never really a problem in the first place; it was a test-specific race condition.
• If a Node already has a TreasureMap, exactly matching one that Alice is trying to PUT, the Node returns a response with status code 309. Fixes REST response when a TreasureMap already exists #341.
• Changed Alice's methods around Arrangement proposal to propose_arrangement instead of consider_arrangement. Fixes Alice doesn't consider arrangements. She proposes them. #1924.
• The lifecycle of test Ursulas is now much more carefully controlled, with each fixture that creates test Ursulas now responsible for removing them.