Update maze transformer: modified RL components and added experiment configs

feat: rl_mem — heap long/short-term memory (short=local window; long=importance top-M memory heap); vectorized top-k causal mask + null-memory NaN guard; launch/eval configs + 7 tests

feat: unified experiment launcher — every experiment = one JSON config (experiments/*.json); launch.py dispatches rl_nav/rl_rnn/rl_ctx/evaluate/random; 11 configs + 11 tests