fix: preserve zero LlamaIndex chunk overlap

[VectorPeak]

Fix LlamaIndex chunk geometry so an explicit chunk_overlap of 0 is preserved instead of falling back to DeepTutor's default overlap.

Description

chunk_geometry() previously used settings.get("chunk_overlap", 50) or 50, which treated a valid configured 0 as missing and silently returned 50. The runtime settings layer and UI already allow chunk_overlap=0, so this made the indexing runtime ignore a user's attempt to disable overlap.

This change distinguishes None from other falsy values when applying defaults, preserving explicit zero overlap while keeping the existing 512/50 fallback for missing or unreadable settings.

Related Issues

• No linked issue. This is a small LlamaIndex runtime configuration bug fix.
• Related but non-overlapping: feat: embed local Markdown image references as data URIs during indexing #525 changes Markdown image preprocessing in the LlamaIndex document loader; [Feature Request]: Add HNSW index and Rerank support for scalable retrieval #304 discusses future retrieval scalability via HNSW/reranking. This PR only adjusts LlamaIndex chunk size/overlap configuration, so it should compose cleanly with both.

Module(s) Affected

• agents
• api
• config
• core
• knowledge
• logging
• services
• tools
• utils
• web (Frontend)
• docs (Documentation)
• scripts
• tests
• Other: services/rag/pipelines/llamaindex

Checklist

• I have read and followed the contribution guidelines.
• My code follows the project's coding standards.
• I have run pre-commit run --all-files and fixed any issues.
• I have added relevant tests for my changes.
• I have updated the documentation (if necessary).
• My changes do not introduce any new security vulnerabilities.

Additional Notes

Evidence:

Before this change, this runtime settings payload:

{"chunk_size": 512, "chunk_overlap": 0}

was interpreted by chunk_geometry() as (512, 50) because 0 or 50 selected the fallback. After this change, the same payload returns (512, 0).

Possible call chain / impact:

deeptutor/services/rag/pipelines/llamaindex/config.py
  -> chunk_geometry()
  -> deeptutor/services/rag/pipelines/llamaindex/embedding_adapter.py
  -> Settings.chunk_overlap
  -> deeptutor/services/rag/pipelines/llamaindex/ingestion.py
  -> SentenceSplitter(chunk_overlap=Settings.chunk_overlap)

This PR is scoped to LlamaIndex chunk geometry configuration. It does not change document loading, parser selection, vector indexing, reranking, or chat retrieval UI behavior.

Validation:

• mamba run -n prw-deeptutor-py311 pytest -q tests/services/config/test_llamaindex_settings.py - passed, 5 tests
• mamba run -n prw-deeptutor-py311 python -m ruff check deeptutor/services/rag/pipelines/llamaindex/config.py tests/services/config/test_llamaindex_settings.py - passed
• mamba run -n prw-deeptutor-py311 ruff format --check deeptutor/services/rag/pipelines/llamaindex/config.py tests/services/config/test_llamaindex_settings.py - passed
• git diff --check - passed

Limitation: I did not run the full pre-commit run --all-files; validation was focused on the touched Python files and the relevant configuration tests.