Rarely used or recommended, the Shuffle convenience method was provided for three reasons:

1. During v1 development, a short-term goal was to be able to use Fixie conventions to mimic the behavior of similar frameworks like xUnit and NUnit, so it was added to ease mimicking xUnit's test order behavior specifically. Framework mimicry is no longer an explicit goal, now that the practice served its purpose of uncovering requirements.
2. There was no such routine available in the framework, and having end users write it by hand was subtle enough that it'd risk introducing quiet bugs that would amount to skipping tests. You'd know something was wrong, but the evidence would be far from the cause.
3. Writing your own would involve directly using some Random instance, which has a long history of thread safety issues where the behavior and proper solution varied greatly from .NET Framework, to .NET Core/5, to .NET 6+. Bottling up the recommendation in a convenience method hid the use of Random while using it safely.

.NET 6 introduced Random.Shared for unambiguous thread safety.

.NET 8 introduced Random.Shuffle<T>(...) for the in-place shuffling of a span or array.

The extension method easily misled end users to perform shuffling during the Discovery phase, instead of the Execution phase, which isn't wrong but would easily lead to only shuffling test methods within a class, for instance, while the test class order remained fixed, without being apparent to the developer. Users will be better off altering test order in their Execution phase where the full impact of their implementation will be more clear.

Now that we target .NET 8+, we have no reason to continue maintaining or suggesting this code. Users can call Random.Shared.Shuffle(...) and ideally in their IExecution implementation.