Untangle the testcases environments

[f-f]

Right now the tests are "coupled": basically the whole test setup is a continuous sequence of commands (i.e. the subsequent commands "innherit" the environment from the previous ones) and we test that those combinations work. The reason is that sometimes it makes sense to see that a command executes correctly and "fixes" a problem introduced by the previous one, and so on.

We should:

• decouple the test cases so that they are responsible for creating their own environment
• allow them to run in parallel, by running every test case in a separate temp directory