feat: make file vectorization strategy configurable

[ningfeemic-dev]

Summary

Refs #857.

This PR makes text file vectorization strategy configurable to reduce embedding oversize failures on long text inputs.

What changed

• add embedding.text_source with supported values:
  • summary_first
  • summary_only
  • content_only
• add embedding.max_text_chars to cap raw text sent to embeddings
• update vectorize_file() to respect the new config
• add minimal validation/unit tests for config and strategy behavior

Why

Current upstream behavior still defaults to full-text embedding for text files. In real deployments using OpenAI-compatible embedding backends, this can trigger repeated oversize failures like input (...) is too large to process, reducing indexing completeness and operational stability.

Making this configurable gives operators a safer, backward-compatible way to balance:

• stability
• indexing completeness
• retrieval quality

Notes

• This PR intentionally keeps the scope small.
• It does not introduce more complex chunking logic yet.
• It focuses on configurable text source selection plus max raw-text length control.

Validation

• source files and tests pass py_compile
• targeted runtime-style validation in the current environment was limited by missing test dependencies for the upstream repo checkout, so this PR includes focused unit tests for the new config/strategy paths

[CLAassistant]

All committers have signed the CLA.

[github-actions]

Failed to generate code suggestions for PR