Expose split_by_character parameter in HTTP API /documents/text endpoint

[gn1024]

Problem

The Python API LightRAG.ainsert() supports split_by_character and split_by_character_only parameters for custom chunk splitting. However, the HTTP API endpoints /documents/text and /documents/texts in lightrag/api/document_routes.py do not expose these parameters — they are not included in the Pydantic request models and not passed through to ainsert().

This forces HTTP API users to rely solely on the built-in token-based chunker, even when they have pre-chunked content with a known separator.

Use Case

We pre-chunk documents with a semantic chunker (heading-aware, with breadcrumbs and atomic blocks) before sending to LightRAG. We join chunks with a unique separator and want LightRAG to split on it, preserving our chunk boundaries as-is.

Without split_by_character in the HTTP API, the only options are:

1. Send each chunk as a separate document (/documents/texts with N items) — creates N doc_ids per file, breaks deletion, deduplication, and doc_status tracking.
2. Use the Python API directly — not possible when LightRAG runs as a separate service.

Proposed Change

Add split_by_character and split_by_character_only fields to InsertTextRequest and InsertTextsRequest in document_routes.py, and pass them through to rag.ainsert().

InsertTextRequest

class InsertTextRequest(BaseModel):
    text: str
    # ... existing fields ...
    split_by_character: Optional[str] = Field(
        default=None,
        description="Character(s) to split the text on instead of token-based chunking",
    )
    split_by_character_only: bool = Field(
        default=False,
        description="If True, split only on split_by_character without token-based fallback",
    )

Route handler

await rag.ainsert(
    request.text,
    split_by_character=request.split_by_character,
    split_by_character_only=request.split_by_character_only,
)

Same for InsertTextsRequest / /documents/texts.

Notes

• Fully backward compatible — both fields are optional with defaults matching current behavior.
• We have a working patch in production (LightRAG v1.4.14) and can submit a PR.