feat: env var overrides for extraction and query prompts

[NaveRazy-Rhino]

Description

Add environment variable overrides for the key prompts in lightrag/prompt.py, allowing users to customize entity extraction, summarization, and query prompts without modifying source code.

This is a minimal, non-breaking bridge until the full prompt template management system (#2652) lands. It follows the same pattern LightRAG already uses for ENTITY_TYPES and SUMMARY_LANGUAGE.

Related Issues

• Closes Custom Entity Types and Prompts #308 — Custom Entity Types and Prompts
• Closes Custom prompt for entity extration #1672 — Custom prompt for entity extraction
• Related to [Feature Request]: Prompt Template Management with Groups and Versioning #2623 — Prompt Template Management (this PR is a lightweight stopgap)
• Related to Feat: Support Prompt Template Management with Types and Versioning #2652 — Full prompt template system (this PR complements, not competes)

Changes Made

lightrag/prompt.py (42 lines added):

• Added import os, json
• Added env var override loop at module bottom: if an env var is set, it replaces the corresponding PROMPTS entry; otherwise the hardcoded default is preserved
• 6 string prompt overrides: ENTITY_EXTRACTION_SYSTEM_PROMPT, ENTITY_EXTRACTION_USER_PROMPT, SUMMARIZE_ENTITY_DESCRIPTIONS_PROMPT, RAG_RESPONSE_PROMPT, NAIVE_RAG_RESPONSE_PROMPT, KEYWORDS_EXTRACTION_PROMPT
• 2 list prompt overrides (JSON arrays): ENTITY_EXTRACTION_EXAMPLES, KEYWORDS_EXTRACTION_EXAMPLES

env.example (15 lines added):

• Documented all new env vars with usage notes, placed next to ENTITY_TYPES

Why This Matters

Domain-specific use cases — code analysis, legal, medical, DevOps — need different extraction prompts. For example, the default prompt title-cases all entity names (Split And Save Layers), which is wrong for code identifiers that must preserve snake_case. Currently the only options are editing prompt.py directly or monkey-patching at runtime.

This PR lets users set ENTITY_EXTRACTION_SYSTEM_PROMPT in .env and get custom extraction behavior with zero code changes, zero new dependencies, and zero risk to existing users.

Example Usage

# In .env — customize for code analysis
ENTITY_TYPES='["Function", "Class", "Module", "ConfigKey", "Service", "Concept"]'
ENTITY_EXTRACTION_SYSTEM_PROMPT='---Role---\nYou are a Code Knowledge Graph Specialist...\n(custom prompt preserving snake_case identifiers)'

Checklist

• Changes tested locally
• Pre-commit checks pass (pre-commit run --all-files)
• Documentation updated (env.example)
• Zero breaking changes — all overrides are optional with existing defaults
• No new dependencies

Additional Notes

This is intentionally minimal. The full prompt template system in #2652 (Jinja2, git-backed, WebUI editor) is the right long-term solution. This PR is a 57-line bridge that unblocks the community today — it can be deprecated when #2652 merges.