A few `DAG.__getitem__` "edge" cases.

[thorwhalen]

It seems like order get's screwed up. See below that dag[:'b c'] gives the wrong anwer (the same thatdag[:'c b]` gives).

Also, in dag[:'b d'] one would expect to get the value of b and the value of d, but we only get d..

[thorwhalen]

See also #60

[thorwhalen]

Confirmed still reproducing on master (0cb2609), both symptoms:

```python

dag: a -> f -> b, b -> g -> c, b -> h -> d

dag:'b c' # 4 (== dag:'c b'; order of 'b c' vs 'c b' has no effect, and neither
# returns a (b, c) tuple -- just c's value)
dag:'b d' # 1 (== h(f(1)) only; b's own value is dropped, not returned alongside d)
```

So both parts still hold: order in the multi-name string doesn't affect output ordering/shape, and asking for 'b d' (a non-terminal + a terminal) silently drops the non-terminal's value instead of returning both. Labeling bug; leaving as a plan item since fixing the output shape (tuple ordering, mixed terminal/non-terminal requests) touches DAG.__getitem__'s core semantics.