Our handling of `&`, modeled after `.`, was dead wrong.  The point is
to capture all the sums that appear at the top-level, so `(a?)[bc]` is
split in `b⊕c⊕a(b+c)`.  This did require to splut `[bc]` because `a?`
is nullable.

However, doing the same with `&` splits `(a?)&[bc]` into
`b⊕c⊕a&(b+c)`, which is obviously wrong (e.g., the latter accepts `b`
but not the former).

Don't try to split extended operators.  Eventually, we might decide to
distribute from the left to the right, in which case `(a?)&[bc]`
should be split in `\e&[bc] ⊕ a&[bc]`, but not what we had currently.

See issue #149 about that.

* vcsn/algos/split.hh: Avoid recreating equivalent expressions:
just reuse the expression itself when there is nothing to do.
* tests/bin/test.py (shortest): New.
* tests/python/split.py (check): Check that the split expression is
equivalent to the expression itself.
Check extended operators.