Fix Command stubtest errors

[jorenham]

Following #787, here's the next batch of stubtest fixes. This time for the sh.Command class.

before:

$ stubtest sh
<errors omitted for sanity>
Found 34 errors (checked 1 module)

after:

$ stubtest sh
<errors omitted for sanity>
Found 25 errors (checked 1 module)

There's one place where I deviated from runtime, which is for the return type of __class_getitem__. At runtime it directly returns cls, but it's better to have it return a types.GenericAlias so that e.g. print(Command[str]) will be printed as Command[str]. Currently it also allows nonsensical expressions like Command[str][str][str], which won't be possible anymore if we'dd change that to return a GenericAlias.
Do you want me to take care of that as well? And if so, should I put it in here, or in a separate PR?

[amoffat]

Sure you can put it in this PR if it's easier. Also, given that I had never heard about stubtest before, is it reasonable to have another lint CI step for it?

[jorenham]

given that I had never heard about stubtest before,

FWIW; It's used by typeshed itself, and we also use it at NumPy, for example, so it's pretty decent stuff.

is it reasonable to have another lint CI step for it?

Yea that's indeed what I was planning on after fixing the errors :)