fix(embedding): drop degenerate tiny chunks before embedding

[lfnovo]

Summary

Fixes #764. URL source ingestion was crashing on complex HTML pages (Wikipedia, Project Gutenberg) with:

Failed to generate embeddings: float() argument must be a string or a real number, not 'NoneType'

Two interacting bugs caused this:

1. Our chunker emits degenerate chunks. LangChain's HTMLHeaderTextSplitter on complex pages can produce single-character or punctuation-only chunks (e.g. "."). Our existing [c.strip() for c in chunks if c and c.strip()] filter only drops empty/whitespace chunks, so these survived.
2. Esperanto crashes on null embeddings returned by llama.cpp's OpenAI-compatible endpoint for tiny inputs in batch mode. Defensive issue filed upstream: Embedding providers crash on null values returned by OpenAI-compatible endpoints esperanto#119.

This PR fixes our side: chunk_text() now filters chunks below OPEN_NOTEBOOK_MIN_CHUNK_SIZE tokens (default 5) before returning. The filter is bypassed if it would empty the result list, so legitimately short documents are preserved.

Changes

• open_notebook/utils/chunking.py: new MIN_CHUNK_SIZE constant + filter step in chunk_text().
• tests/test_chunking.py: 2 new tests covering the filter behavior and the empty-result safeguard.
• docs/5-CONFIGURATION/environment-reference.md: documents the new env var.
• CHANGELOG.md: Unreleased entry.

Test plan

• uv run pytest tests/test_chunking.py tests/test_embedding.py — 52 passed
• Manual: ingest https://en.wikipedia.org/wiki/Martin_Luther via URL with a llama.cpp embedding endpoint and confirm processing completes

[cubic-dev-ai]

No issues found across 4 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.