fix(rag): split oversized markdown header chunks

[wolfkill]

Summary

• Fixes [Bug]RAG markdown default chunk strategy fails on large embedded YAML/TOML code blocks (single oversize chunk) #3030.
• Keeps markdown header-based grouping, but applies a size-bound second pass when an aggregated markdown header chunk exceeds chunk_size.
• Preserves header metadata on each generated sub-chunk.
• Avoids introducing an overlap validation regression when a caller sets a small chunk_size without overriding the splitter's default overlap.

Why

Markdown documents can contain large fenced YAML/TOML/config blocks under a single header. The header splitter previously returned that whole section as one chunk, which can exceed embedding provider input limits even when chunk_size is configured.

Tests

• Red test before implementation:
  • .venv/bin/python -m pytest packages/dbgpt-core/src/dbgpt/rag/text_splitter/tests/test_splitters.py::test_md_header_text_splitter_splits_oversized_code_block -q
  • Failed as expected because only one oversized chunk was produced.
• .venv/bin/python -m pytest packages/dbgpt-core/src/dbgpt/rag/text_splitter/tests/test_splitters.py::test_md_header_text_splitter_splits_oversized_code_block -q
• .venv/bin/python -m pytest packages/dbgpt-core/src/dbgpt/rag/text_splitter/tests/test_splitters.py packages/dbgpt-ext/src/dbgpt_ext/rag/knowledge/tests/test_markdown.py -q
• Manual default-path check with MarkdownKnowledge + ChunkManager + ChunkParameters(chunk_strategy="Automatic", chunk_size=128, chunk_overlap=16):
  • Result: 17 chunks, max length 127, header/source metadata preserved.
• .venv/bin/python -m ruff check packages/dbgpt-core/src/dbgpt/rag/text_splitter/text_splitter.py packages/dbgpt-core/src/dbgpt/rag/text_splitter/tests/test_splitters.py
• .venv/bin/python -m ruff format --check packages/dbgpt-core/src/dbgpt/rag/text_splitter/text_splitter.py packages/dbgpt-core/src/dbgpt/rag/text_splitter/tests/test_splitters.py
• .venv/bin/python -m compileall -q packages/dbgpt-core/src/dbgpt/rag/text_splitter/text_splitter.py packages/dbgpt-core/src/dbgpt/rag/text_splitter/tests/test_splitters.py
• git diff --check

[wolfkill]

Follow-up update:

• Rebased this branch onto the latest origin/main after fix(rag): use size chunking by default for markdown knowledge (Fixes #3030) #3033 added markdown size chunking by default.
• Resolved the splitter conflict by keeping the upstream fallback split behavior and only clamping the fallback overlap to avoid introducing a chunk_overlap > chunk_size error for direct MarkdownHeaderTextSplitter(chunk_size=...) usage.
• Also formatted react_parser.py and test_react_parser.py because the repository CI runs full-package make fmt-check, and those files were the current format blocker.

Local verification:

• PYTHONPATH=packages/dbgpt-core/src .venv/bin/python -m pytest packages/dbgpt-core/src/dbgpt/rag/text_splitter/tests/test_splitters.py -q -> 5 passed
• PYTHONPATH=packages/dbgpt-core/src .venv/bin/python -m pytest packages/dbgpt-core/src/dbgpt/agent/util/tests/test_react_parser.py -q -> 24 passed
• make fmt-check -> passed
• PYTHONPATH=packages/dbgpt-core/src .venv/bin/python -m compileall -q packages/dbgpt-core/src/dbgpt/rag/text_splitter/text_splitter.py packages/dbgpt-core/src/dbgpt/rag/text_splitter/tests/test_splitters.py packages/dbgpt-core/src/dbgpt/agent/util/react_parser.py packages/dbgpt-core/src/dbgpt/agent/util/tests/test_react_parser.py -> passed
• git diff --check -> passed

GitHub CI is now green on this PR.