support 'yield' statement/expression in parser to provide at least a more meaningful error message for now

- coercion code cannot currently be taken out of the conditional as the temp release of the coercion code happens straight away

- in the case of an unpacking error, no assignment must take place
  => do the complete unpacking first, then the assignment
- a tiny bit slower for tuples (<5%)
- shorter, much more readable code

enable iter-dict optimisation also for a plain 'for x in dict', assign dict ref to temp var before entering the loop to avoid re-assignment problems

new transform that converts for-in-dict.iter*() into a while-loop over PyDict_Next(), which makes the loop 30-50% faster