fix(pu): index sampled target policies by slot

[puyuan1996]

Closes #486

Summary

• Keep base MuZero target policy assignment indexed by real legal action ids.
• Override sampled MuZero/EfficientZero target policy assignment to use sampled-action slots instead of sparse environment action ids.
• Route both non-reanalyzed and sampled reanalyzed target policy construction through the same assignment helper.
• Add a regression test for sparse legal action ids with SampledMuZeroGameBuffer and SampledEfficientZeroGameBuffer.

Tests

• /mnt/shared-storage-user/puyuan/lz/bin/python -m pytest lzero/mcts/tests/test_game_buffer.py -q
• /mnt/shared-storage-user/puyuan/lz/bin/python -m py_compile lzero/mcts/buffer/game_buffer_muzero.py lzero/mcts/buffer/game_buffer_sampled_muzero.py lzero/mcts/buffer/game_buffer_sampled_efficientzero.py lzero/mcts/tests/test_game_buffer.py
• /mnt/shared-storage-user/puyuan/lz/bin/python -m pytest lzero/model/tests/test_sampled_efficientzero_model.py -q