Distinguish empty call defaults from comprehension accumulators

[EiNSTeiN-]

I am opening this PR to fix nested comprehension reconstruction when an iterator-producing call takes an empty container as a default argument. The code below currently loses its comprehension structure and is reconstructed correctly with this change:

def select(rows):
    return [field for row in rows for field in row.get("values", [])]

An empty list or map preceding an iterator need not be a new comprehension's result container: it may already have been consumed as a method argument. The prologue detector currently accepts an earlier empty BUILD_LIST, BUILD_SET or BUILD_MAP without distinguishing those uses. The fix requires the modern accumulator's immediate SWAP 2 placement before treating it as a new comprehension level.

The same distinction matters for set and dictionary results:

{field for row in rows for field in row.get("values", {})}
{field: len(field) for row in rows for field in row.get("values", {})}

Separately nested comprehensions must keep their own accumulator, including:

[[field for field in row.get("values", [])] for row in rows]
[{field for field in row.get("values", {})} for row in rows]
[{field: len(field) for field in row.get("values", {})} for row in rows]

Direct row["values"], a call without a default, and a tuple default retain their existing interpretation. The surrounding function's saved row and field bindings must also survive comprehension cleanup.

This accumulator layout begins with comprehension inlining in Python 3.12. The additional SWAP 2 requirement therefore applies from 3.12, preserving the older prologue handling.