fix: walk result expression value to prevent false positive in UnusedVariables

[frankentini]

Summary

Fixes #226.

unused_assignments() breaks out of its loop upon encountering the result = ... assignment. However, it did so before walking the assignment's value for variable references. This caused any variable only referenced within the final result expression (e.g. via BINGET/memo-based sharing, or as part of a complex result tuple/dict) to be incorrectly flagged as unused.

Root Cause

In Interpreter.unused_assignments() (fickle.py), the loop processed statements sequentially:

if statement.targets[0].id == "result":
    break  # <-- exits WITHOUT walking statement.value

Variables referenced exclusively in the result expression were never added to the used set, so they appeared in defined - used and were falsely reported as suspicious.

Fix

Before breaking, walk the result assignment's value to record all variable references:

if statement.targets[0].id == "result":
    for node in ast.walk(statement.value):
        if isinstance(node, ast.Name):
            used.add(node.id)
    break

Test

Added test_unused_variables_no_false_positive_for_result_refs which crafts a minimal pickle where a REDUCE-created variable is referenced only in the result tuple via BINGET. Without the fix, this variable is falsely flagged; with the fix, it correctly reports zero unused variables.

This directly reproduces the pattern seen in the scanpy pickle from #226, where variables created by numpy.core.multiarray.scalar REDUCE calls end up referenced only within result-level structures.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}