automatically test for out of memory conditions when `std.testing.allocator` is used

[andrewrk]

Zig already has checkAllAllocationFailures for doing this, however it is opt-in.

I propose to make the test runner enable this kind of testing by default. When it makes an allocation fail, it allows error.OutOfMemory to be returned from the unit test, but still checks for memory leaks.

A unit test could opt-out of this by not using std.testing.allocator or by setting a flag (i.e. std.testing.skipAllocationFailureChecking();).

[squeek502]

This possibility is something I mentioned at the end of this post I wrote about checkAllAllocationFailures. Some potential wrinkles to consider:

• This will increase the time it takes to run tests that allocate, since each test will be run N+1 times where N is the number of allocations that occur in the test
• Tests may use code that includes a strategy for handling OutOfMemory (i.e. somewhere along the line they catch error.OutOfMemory and it won't make its way all the way back to the caller). With checkAllAllocationFailures, error.SwallowedOutOfMemoryError is returned in this scenario which users can catch and ignore if they know it's intentional. With test runner integration, tests might need to be able to communicate that not returning OutOfMemory is okay.
  • I think this situation is rarer now than it was when I originally wrote checkAllAllocationFailures because of Enable FailingAllocator to fail on resize #17007. In saying that, checkAllAllocationFailures currently doesn't induce resize failures, so that might be something to look into as well.
• There is a possible situation where the FailingAllocator is intending to induce a failure, but the allocation index is never reached. This is indicative of a non-deterministic amount of allocations in the test, and means that the checkAllAllocationFailures strategy won't be reliable for that test case. Currently, checkAllAllocationFailures returns error.NondeterministicMemoryUsage in this scenario.