Memory leak when calling list(...) on OrderedDict

Description

There is a memory leak when calling the list function with an OrderedDict. Even after the runtime has been shut down, objects of type PythonGetMemberBinder, PythonType, FunctionDefinition, FunctionCode etc. remain referenced and will not be garbage collected.

Steps to Reproduce

(Repro project available here: https://github.com/spkl/IronPythonMemoryRepro)

1. In a .NET 6 program, create a ScriptEngine and ScriptScope.
2. Create a ScriptSource from a file containing the following code:

import collections

def MyMethod(dummy):
    od = collections.OrderedDict()
    for i in range(100):
        od[i] = []
    
    for v in list(od.values()):
        continue

    return dummy

3. Execute the ScriptSource.
4. Call the MyMethod function.
5. Shut down the engine.
6. Repeat this 50 times.

Expected behavior: The memory footprint is stable. (For reference: After the first execution, it is ~50MB).

Actual behavior: The memory footprint exceeds 300 MB after 50 repetitions. Forcing a garbage collection does not change it significantly. (With a larger, more complex example, we have observed ~600 MB of additional, uncollectable memory after 30 script executions.)

Workaround: Instead of for v in list(od.values()):, use for v in [x for x in od.values()] or for v in od.values():.

Versions

The issue is reproducible with both 3.4.0 and 3.4.1.

Start by running the linked IronPythonMemoryRepro with the .NET 6 steps and compare list(od.values()) with the documented workarounds. Trace references to PythonGetMemberBinder, PythonType, FunctionDefinition, and FunctionCode after engine shutdown; done means repeated executions keep memory stable and those objects can be collected.