examples: add clocked_perft.py (timed perft, canonical node tables)

[FreakyAdy]

Adds examples/clocked_perft.py -- a self-contained, wall-clocked perft
using only the python-chess move generator (no engine, no NN, no search
package).

Verified locally (this machine, canonical tables -- exit 0):

start        d1=20   d2=400    d3=8902     d4=197281     # canonical
KiwiPete     d1=48   d2=2039   d3=97862    d4=4085603    # canonical

The wall-clock + nodes/sec output is exactly the budget-vs-latency lever
that RL-for-Universal-Compiler-Optimization and chessy benchmark when
tuning move-generator throughput on low-end hardware.

Run:

python examples/clocked_perft.py
python examples/clocked_perft.py "<FEN>" 4

Output (one parseable line):

total_nodes_per_s=<x> total_wall_s=<s> total_nodes=<n> perft={1:a,2:b,...}