examples: make 08_python_subport run a real Python node

The example was named "python subport" but its graph was entirely C++
(ProduceNode -> DoubleItNode); Python only tapped the output, leaving a
dangling "#todo: return value to network".

Rewrite so the only node in the graph is a pure-Python py_triple, driven
from both ends via the subport taps: net.write() injects inputs and
net.read() pulls results back, closing the round trip. Also del the
network at the end so its callable cycle is reclaimed deterministically.

examples/08_python_subport/example.py

@@ -1,38 +1,55 @@
 """
-08_python_subport — tap a C++ node's output from Python using net.read().
+08_python_subport — drive a *Python* node from Python via write()/read() taps.
 
 Graph:
-  [ProduceNode] --int--> [DoubleItNode] --int--> (tapped by net.read())
+  (fed by net.write()) --int--> [py_triple] --int--> (tapped by net.read())
 
-The sink is Python: instead of connecting a PrintItNode, we call net.read()
-to pull values out of DoubleItNode's output directly into Python.
-We also demonstrate net.write() by injecting a value into DoubleItNode's input.
+Unlike 07, there is no C++ source or sink here: the only node in the network is
+a pure-Python function, py_triple. Python plays *both* the producer and the
+consumer by using the subport taps:
+
+  * net.write("py", 0, v)  injects v into py_triple's input   (Python -> network)
+  * net.read("py", 0)      pulls py_triple's output back out   (network -> Python)
+
+This closes the loop the old version left as a "#todo": a value flows from
+Python, through a Python node running inside the network, and back to Python.
 """
 
 import sys
-import time
-import threading
-sys.path.insert(0, "build/python")
+sys.path.insert(0, "build/python")  # for `python examples/.../example.py` from repo root
 
 import kpn_python as kpn
 
+
+def py_triple(x: int) -> int:
+    return x * 3
+
+
 net = kpn.Network()
 
-net.add("src", kpn.make_produce())
-net.add("dbl", kpn.make_double_it())
+# The whole network is a single Python node with a tapped input and output.
+net.add_node("py", py_triple, inputs=["int"], outputs=["int"])
 
-net.connect("src", 0, "dbl", 0)
 net.build()
 net.start()
 
-# Collect a few values from DoubleItNode's output via Python tap
+# Push values in from Python and read the Python node's results back out.
+inputs = [1, 2, 7, 10, 100]
 results = []
-for _ in range(5):
-    val = net.read("dbl", 0)
-    results.append(val)
+for v in inputs:
+    net.write("py", 0, v)          # Python -> py_triple input
+    results.append(net.read("py", 0))  # py_triple output -> Python
 
 net.stop()
 
-print("values read from C++ DoubleItNode output:", results)
-assert all(v == 84 for v in results), f"expected all 84, got {results}"
-print("all correct (42 * 2 = 84)")
+print("inputs written from Python: ", inputs)
+print("outputs read from py_triple:", results)
+
+expected = [v * 3 for v in inputs]
+assert results == expected, f"expected {expected}, got {results}"
+print("all correct (x * 3 computed by a Python node inside the network)")
+
+# Drop the network deterministically. The network holds the Python callable,
+# which (via globals) forms a reference cycle; deleting the global breaks it so
+# the network is reclaimed promptly instead of lingering to interpreter exit.
+del net