[WH-2099]

Important

1. Make sure you have read our contribution guidelines
2. Search existing issues and pull requests to confirm this change is not a duplicate
3. Open or identify the issue this pull request resolves or advances
4. Use a Conventional Commits title for this pull request, and mark breaking changes with !
5. Remember that the pull request title will become the squash merge commit message
6. If CLA Assistant prompts you, sign CLA.md in the pull request conversation

Related Issue

Closes #354

Summary

Host agent loops need to retain complete model responses while displaying streamed chunks.
Expose the existing _LLMChunkAccumulator as LLMResultAccumulator so callers can reuse its consume() and to_result() methods for replay and JSON persistence.
Non-stream normalization continues to use the same implementation.

Document one accumulator per invocation and add a public API regression covering mixed content order, an empty content block carrying provider state, exact tool-call arguments, and false-valued assistant state through JSON serialization.
The downstream consumers are in langgenius/dify#44502.

Validation

• The regression fails on main 4389ea1 because the public class is absent; the existing private implementation satisfies its behavior assertions.
• PYTEST_ADDOPTS='-n 4' just test: 995 passed on Python 3.12.
• just check: passed, including lockfile, formatting, lint, types, and import boundaries.
• uv run --python 3.12 pytest -n 0 tests/test_repository_docs.py: 29 passed.
• Separate fresh test, knowledge, and final naming reviews passed.

Checklist

• This pull request links the issue it resolves or advances
• This pull request title follows Conventional Commits, and any breaking change is marked with !
• If CLA Assistant prompted me, I signed CLA.md in the pull request conversation

[Copilot]

🟢 Approval recommended

The narrow API exposure preserves existing behavior, includes focused documentation and regression coverage, and has no unresolved findings.

No open issues 🎉

What changed in this PR

Exposes the existing model-response accumulator as LLMResultAccumulator, allowing hosts to display streamed chunks while retaining a complete response for replay and persistence.

Changes:

• Publishes the accumulator without changing aggregation behavior or non-stream normalization.
• Documents incremental usage and one accumulator per invocation.
• Adds a public API regression covering content order, tool-call arguments, and provider-state preservation through JSON serialization.

File	Description
tests/​model_runtime/​test_llm_result_accumulator.py	Adds public API and JSON round-trip regression coverage.
src/​graphon/​model_runtime/​README.md	Documents incremental accumulation and invocation lifetime.
src/​graphon/​model_runtime/​model_providers/​base/​large_language_model.py	Publishes the accumulator and updates its internal use.

_{Review effort: Balanced}

[laipz8200]

LGTM