Fix ServerArgs when using DP attention

[knukiban]

Motivation

ServerArgs.__post_init__ is not idempotent and is called twice when running bench_offline_throughput, causing chunked_prefill_size to be halved multiple times.
Additionally, https://github.com/sgl-project/sglang/pull/3964/files#diff-700b5118b493d60d7b5994857f5f1e6a7e842ad702392b8ab199945764dfc8edR1144 caused a port collision. (fixed in #4648 though less robustly)

> python -m sglang.bench_offline_throughput --model-path=deepseek-ai/DeepSeek-V2-Lite  --trust-remote-code --enable-dp-attention --dp-size=2 --tp-size=2
INFO 03-20 00:47:17 __init__.py:190] Automatically detected platform cuda.
The following error message 'operation scheduled before its operands' can be ignored.
DP attention is enabled. The chunked prefill size is adjusted to 4096 to avoid MoE kernel issues.
DP attention is enabled. The chunked prefill size is adjusted to 2048 to avoid MoE kernel issues.
server_args=ServerArgs(..., schedule_conservativeness=0.09, ...)

... a bunch of lines ...

zmq.error.ZMQError: Address already in use (addr='tcp://127.0.0.1:30236')

Modifications

• Use the user-specified --chunked-prefill-size or --schedule-conservativeness value if explicitly set. Defaults when running sglang.launch_server are unchanged from main.
• Make ServerArgs.__post_init__() idempotent, so that running sglang.bench_offline_throughput uses the same values as sglang.launch_server
• Fix port incrementing for generating PortArgs when using --enable_dp_attention

Result: This completes and uses default chunked prefill size 4096 on 2 GPUs.

> python -m sglang.bench_offline_throughput --model-path=deepseek-ai/DeepSeek-V2-Lite  --trust-remote-code --enable-dp-attention --dp-size=2 --tp-size=2
INFO 03-20 00:56:45 __init__.py:190] Automatically detected platform cuda.
The following error message 'operation scheduled before its operands' can be ignored.
DP attention is enabled. The chunked prefill size is adjusted to 4096 to avoid MoE kernel issues.
server_args=ServerArgs(..., schedule_conservativeness=0.3, ...)

Checklist

• Format your code according to the Code Formatting with Pre-Commit.
• Add unit tests as outlined in the Running Unit Tests.
• Update documentation / docstrings / example tutorials as needed, according to Writing Documentation.
• Provide throughput / latency benchmark results and accuracy evaluation results as needed, according to Benchmark and Profiling and Accuracy Results.
• For reviewers: If you haven't made any contributions to this PR and are only assisting with merging the main branch, please remove yourself as a co-author when merging the PR.
• Please feel free to join our Slack channel at https://slack.sglang.ai to discuss your PR.

[github-actions]

This pull request has been automatically closed due to inactivity. Please feel free to reopen it if needed.