Fix parens around comment at object-destructuring assignment

[gnprice]

Fixes #575.

We were taking code like the following:

(/**/{foo} = 1)

and wrapping the comment in parens, which is invalid syntax:

(/**/)({foo} = 1)

The symptom appears when using the default parser, but not esprima;
and only when you call the printer directly on the AST node for the
expression, not the whole program.

The bug here was that in needsParens, we check to see if node is
an AssignmentExpression with ObjectPattern, and if so we say that
we do need parens... but we were making that check at the very top
of the function, before we'd checked that node was actually the
thing we wanted to be operating on at all.

If in fact the path currently points to a comment -- as it can
when needsParens gets called by the printer invoked by the
print(commentPath) in printLeadingComment or printTrailingComment
-- then the comment doesn't count as a node, and so getNode will
actually return the parent. If that parent meets the
AssignmentExpression / ObjectPattern condition, then we would
buggily decide that the comment needed parens too.

Fix it by checking that node is the actual top of the stack,
i.e. this.getValue(), before doing anything with node.
Also add a regression test.