Add ability to stop after some number of failed tests

[nfachan]

TAC

• This should be a configuration value. A user could want to set this in their configuration.
• The configuration value should be called stop-after. I'm open to other suggestions. I prefer imperative verbs like "stop", but I'll admit it's not obvious what we're stopping after. We could consider something like max-failures or failure-limit.
• Upon reaching the Nth failure, the test runner should immediately print a summary and exit. This will stop any pending tests. We don't want to wait for outstanding tests.
• The summary message should at a minimum indicate that some tests weren't run. Ideally, we would print more information, but the fact that we may not have built all test binaries complicates things. If we really wanted to go for extra credit, we could print the number of tests that we know haven't been executed, and then possibly add the number of tests that we think will be run, once those test binaries were built.
• We should ensure that cargo test --no-run will immediately exit in such scenarios. Presumably it will get a SIGPIPE eventually, but we may want to accelerate the process.

Definition of Done

• CHANGELOG.md updated.
• Documentation in doc/book/head updated.
• Issue is in the correct milestone.

[nfachan]

This is complicated by build failures. We should think this through a little bit.

cargo test and cargo nextest have --fail-fast and --no-fail-fast.

[nfachan]

Yeah, that's definitely a complexity. I'm wondering if we want to have separate options for the two, and then maybe a convenience option that can control both of them at the same time. When cargo metest runs cargo, it could pass --fail-fast or --no-fail-fast, right?

[nfachan]

I don't think we have to worry about build failures. The current behavior is to immediately stop when the build fails, which is what I think we want to continue to do.