Set forkEnv to an empty Map

[lolgab]

Since os-lib propagates the system environment by default when it runs subprocesses, setting forkEnv to the system environment, is redundant, but also confusing, since removing a key from the Map doesn't stop mill to pass the environment variable to the run program. This sets a new default, which is a semantic breaking change, so it should go in Mill 0.12.0, which is a binary compatible but behavior incompatible release.

Pull Request: #3442

[lihaoyi]

What user-visible effect does this have?

[lolgab]

What user-visible effect does this have?

It doesn't make run much different, since it's a T.command so it runs again every time.
I can see a change in behavior if a user used it as a T.input to get the environment, because now it's a fixed T {} of an empty map, but that's not the "correct" use of this target. It is still safe to change it in 0.12.0 but it doesn't seem visible in the "normal" use case.

[lihaoyi]

Sounds good!