Normative: Support optional chain in LHS of assignment

[nicolo-ribaudo]

Motivation: https://gist.github.com/nicolo-ribaudo/d264e424b618e7deaeca1d6e4f16a7c0

Description of the changes:

1. expr1?.b.c = expr2 is equivalent to expr1 == null ? undefined : (expr1.b.c = expr2)
   • short-circuiting the RHS matches how people currently write that code, i.e. if (expr1) expr1.b.c = expr2;.
2. expr1?.b.c += expr2 is equivalent to expr1 == null ? undefined : (expr1.b.c += expr2)
3. expr1?.b.c ??= expr2 is equivalent to expr1 == null ? undefined : (expr1.b.c ??= expr2)
4. (a?.b) = c throws for nullish a, similarly to how (a?.b).c throws for nullish a.
5. ++expr1?.b.c, equivalent to expr1 == null ? undefined : ++expr1.c.d, would be very confusing, because the ++ operator can be skipped depending on an optional chain that appears after it. As such, it is still disallowed.
   1. For consistency also expr1?.c.b++ is still disallowed, even if it doesn't have the same problem.
6. optional chains are still disallowed inside destructuring expressions: in { x: expr1?.b.c } = obj it's not clear if the obj.x getter should be triggered or not.
7. optional chains are still disallowed in the head of for-in/of loops: for (expr1?.c.d of arr), because it wouldn't short-circuit the evaluation of arr

5.1, 6 and 7 are the points I'm most unsure about, open for discussion in plenary.