Supports DP+TP+EP hybrid parallel deployment strategy

[carryyu]

Supports DP+TP+EP hybrid parallel deployment strategy.

1. Expert_parallel_size = data_parallel_size * tensor_parallel_size when enable-expert-parallel.
2. Deprecated event_loop_ep.
3. Save_output when tensor_parallel_rank == 0.
4. Most distributed operations need to be restricted in tp_group.
5. Weight splitting logic, only splitting the attention part when ep parallelization is turned on.
6. If bias exists in RowParallelLinear, it needs to be divided by tensor_parallel_size to prevent repeated accumulation.
7. To ensure balanced EP communication, MoE Layer will split the input into tensor_parallel_size parts, and then perform AllGather when TP and EP are both enabled.

TODO

1. Hardware backends other than GPUs need to be adapted.
2. Distributed operations require support for 0-dimensional tensors.
3. AllGather will hang when data shapes are different on multi-ranks.
4. Custom_all_reduce supports specified tp_group.
5. Functional and performance verification in various scenarios, such as MTP, ChunkPrefill, and PromptCache.
6. Some details may not be fully covered in this PR.

[paddle-bot]

Thanks for your contribution!

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (develop@c43a4be). Learn more about missing BASE report.

Additional details and impacted files

@@            Coverage Diff             @@
##             develop    #3489   +/-   ##
==========================================
  Coverage           ?   31.44%           
==========================================
  Files              ?       13           
  Lines              ?      159           
  Branches           ?       29           
==========================================
  Hits               ?       50           
  Misses             ?       96           
  Partials           ?       13           

Flag	Coverage Δ
diff	31.44% <ø> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.