[Docs] Fix default values and options in Ascend server arguments documentation

[xdtbynd]

Motivation

This pull request updates the documentation for server arguments to accurately reflect the current defaults, available options, and support status for various features in the Ascend backend (A2, A3). Several descriptions were outdated or contained placeholder information. This change ensures users have correct configuration guidance and are aware of features still in planning.

Key issues addressed:

• Correcting the default value for --speculative-draft-load-format.
• Updating the --tool-call-parser options to match the currently implemented parsers for Ascend.
• Clarifying the planned but unsupported status of --disaggregation-decode-enable-offload-kvcache.
• Removing duplicate table entries in the HTTP server section.

Modifications

The following sections in the server arguments documentation have been updated:

1. HTTP Server:

   • Removed a duplicate table block.
   • Standardized the --grpc-mode server supported description to explicitly state Planned.

2. API related:

   • --tool-call-parser: Updated the list of valid options to llama3, pythonic, qwen, and qwen3_coder to reflect the currently supported parsers on the Ascend backend.
   • Minor formatting adjustments to the --reasoning-parser options.

3. Speculative decoding:

   • --speculative-draft-load-format: Changed the default value from None to auto.
   • --speculative-draft-model-revision: Expanded the "Options" description to include specific examples (branch name, tag name, commit id).

4. Disaggregation:

   • --disaggregation-decode-enable-offload-kvcache: Updated the server supported column to server supported description to explicitly state Planned to accurately indicate the current lack of support for KV cache offloading on Ascend.

Accuracy Tests

N/A - This PR only updates documentation and does not modify any model forward code or kernel logic.

Speed Tests and Profiling

N/A - This PR only updates documentation and does not impact inference speed.

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review and Merge Process

1. Ping Merge Oncalls to start the process. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • Common commands include /tag-and-rerun-ci, /tag-run-ci-label, /rerun-failed-ci
4. After green CI and required approvals, ask Merge Oncalls or people with Write permission to merge the PR.

[gemini-code-assist]

Code Review

This pull request updates the documentation for Ascend NPU support features by refining the available options and default values for several server arguments, including --grpc-mode, --tool-call-parser, and speculative decoding settings. A typo was identified in the description for the --disaggregation-decode-enable-offload-kvcache argument.