[WIP] Add bitwise parity test for MoE EP

[wwwjn]

Stack from ghstack (oldest at bottom):

• -> [WIP] Add bitwise parity test for MoE EP #3172
• [WIP]Enable DP-to-EP for MoE inference #3171
• [rl] Enable TP2EP for MoE inference in vLLM wrapper #3142
• [MoE] Pad token count to a multiple of sp_size in AllToAllTokenDispatcher #3193

Add rl_grpo_qwen3_moe_debug_ep_batch_invariant config and
TestBitwiseParityMoEEP test class for verifying bitwise identity
between trainer prefill and vLLM generator prefill with MoE EP.

Config: TP=4, EP=4 on 4 GPUs, deterministic mode, compile disabled.
Uses /tmp/debug_moe_ckpt by default (override with MOE_HF_ASSETS_PATH).

Run with:
python scripts/rl/create_debug_moe_ckpt.py
torchrun --nproc_per_node=4 -m pytest \
torchtitan/experiments/rl/tests/test_bitwise_parity.py::TestBitwiseParityMoEEP

Note: bitwise parity not yet achieved — current max_delta ~2e-2.
Likely causes: batch-invariant mode disabled (requires batch_invariant_ops),
attention impl differences (varlen vs paged), MoE token routing order.