Intermittent cabal-install integration test failure on Windows

I noticed a weird failure while testing cabal on AppVeyor:

```
exec
should_run
adds_sandbox_bin_directory_to_path.sh (ignoring stderr): FAIL
Exception: RemoveDirectory "C:\\Users\\appveyor\\AppData\\Local\\Temp\\1\\cabal-install-test-1924\\should_run\\.cabal-sandbox\\bin": unsatisified constraints (The directory is not empty.)
auto_configures_on_exec.sh (ignoring stderr): OK (0.50s)
can_run_executables_installed_in_sandbox.sh (ignoring stderr): OK (8.76s)
```

Here is the full log: https://ci.appveyor.com/project/grayjay/cabal/build/1.0.26

The test passed in four other AppVeyor builds, and I haven't been able to reproduce it locally.

EDIT: These tests also need to be enabled in appveyor.yml.

Start with the failing should_run test, adds_sandbox_bin_directory_to_path.sh, the full AppVeyor log, and appveyor.yml, where the issue says these tests need enabling. Compare the failing Windows cleanup behavior with the passing builds and determine what makes the integration test reliable; done means the enabled tests pass consistently on AppVeyor.

Written by the indexing model from the issue text.