test: Add best effort messages to random traffic canister

[stiegerc]

Adds support for best-effort calls on the random traffic canister. All arguments used to run the canister are collected in the Config now (except the random seed) since otherwise the number of update calls and individual parameters would get quite large.
Adds a new test where a local canister makes only best-effort calls, to itself and to a remote subnet. Checks that the local canister can be stopped even if the remote subnet stalls.

To change any parameter you can query the current config, then change it accordingly and send it back to the canister. The seed is excluded since reseeding the rng every time would reset the random sequence.

Additionally the records now contain a timeout if the call is best-effort and the time from making the call to receiving a reply is measured (for all calls).

[alin-at-dfinity]

Checks that the local canister can be stopped even if the remote subnet stalls.

Out of curiosity, why was it that the canister couldn't be stopped (IIRC) the other day?

[oggy-dfin]

LGTM overall, thanks! My main comment is related to the CDK version. Unfortunately, things have been shuffled around a bit in the CDK since you started writing this (for the better, hopefully!) and this would no longer compile with the HEAD of the next CDK version. While we could just pin it to the version you were using and leave it there, it'd suck to have to deal with the version from a random git commit in the future, so I'd propose we clean this up before merging (I don't think it should be a major change).

Furthermore, we should pin the commit in external_crates.bzl, otherwise things might break for someone else who needs to repin the Bazel dependencies (IIUC how the whole Bazel-Rust thing works).

[stiegerc]

Checks that the local canister can be stopped even if the remote subnet stalls.

Out of curiosity, why was it that the canister couldn't be stopped (IIRC) the other day?

I have a function tick() that ticks on both subnets and increments time. When I was trying to simulate one subnet stalling I used the vanilla tick() on StateMachine but this one doesn't increment time. So I just forgot that I have to increment time manually in this case and nothing was timed out.

edit: Well technically tick() on StateMachine does increment time, by 1 nanosecond. So the test would have passed it just would have taken billions of rounds :o)