Support runOrder=balanced with forkCount>1 (without parallel=classes)

Today `runOrder=balanced` looks only at `threadCount` (set by `parallel` or `-DthreadCount`) when it sorts the test classes. If you use `forkCount>1` but not `parallel=classes`, then `threadCount` is `1`. In that case `balanced` only puts the slowest classes first. It does not use the number of forks that run the tests.

I would like `balanced` to also use `forkCount`. The number of parallel test runners is `forkCount × threadCount`. So `balanced` should split the classes over that many runners. Then `balanced` also helps when you use `forkCount>1` without `parallel=classes`.

Idea for the code: send `forkCount` to the provider with a new `forkcount` property (the same way `threadcount` is sent today), and use `threadCount × forkCount` as the number of buckets in `DefaultRunOrderCalculator`.

Trace how threadCount is sent to the provider, then inspect DefaultRunOrderCalculator and the existing balanced run-order calculation. Confirm how forkCount is represented and propagated; done means balanced scheduling uses threadCount × forkCount when forkCount is greater than one without parallel=classes, while existing behavior remains intact.

Written by the indexing model from the issue text.