Add pip-installable reviewer3 CLI for context-aware DOCX paragraph review via OpenAI/Ollama - #1

This PR introduces a Python package and CLI (reviewer3) that reviews scientific .docx files paragraph-by-paragraph using an LLM, then writes feedback back into a new file (*_rv3.docx) without modifying the original.
Feedback generation is context-aware: each paragraph is reviewed with prior summaries, introduced terms/abbreviations, and earlier feedback history.

• Packaging + CLI surface

  • Added pyproject.toml with install metadata and console script entrypoint:
    • reviewer3 = reviewer3.cli:main
  • Added CLI argument handling for input .docx and optional output path.
  • Enforced non-destructive output behavior (input/output path must differ).

• Review orchestration pipeline

  • Added reviewer3/core.py with review_docx(...) flow:
    • iterate non-empty paragraphs
    • request structured LLM feedback per paragraph
    • carry forward running context (previous_summaries, introduced terms, prior feedback)
  • Added env-driven OpenAI client configuration with Ollama-friendly defaults:
    • REVIEWER3_OPENAI_BASE_URL (default http://localhost:11434/v1)
    • REVIEWER3_OPENAI_API_KEY (default ollama)
    • REVIEWER3_OPENAI_MODEL (default gpt-oss:20b)
    • REVIEWER3_OPENAI_TEMPERATURE

• DOCX feedback writing

  • Added comment insertion for constructive questions using python-docx comments API.
  • Added tracked insertion suggestions (w:ins) authored as reviewer3 to represent rewrite suggestions.
  • Default output naming implemented via <input_stem>_rv3.docx.

• Robust parsing + behavior contracts

  • Added structured feedback parsing (parse_feedback) with JSON fallback extraction and no_feedback support.
  • Supports LLM responses that intentionally provide no comment/suggestion for acceptable paragraphs.

• Focused test coverage + docs

  • Added targeted unit tests for:
    • default output naming
    • context propagation across paragraphs
    • comment creation and tracked suggestion insertion in output DOCX
  • Expanded README with tool intent and output behavior.

from pathlib import Path
from reviewer3.core import review_docx, LLMReviewer

reviewer = LLMReviewer.from_env()  # defaults to local Ollama gpt-oss:20b
out = review_docx(
    input_docx=Path("manuscript.docx"),
    output_docx=Path("manuscript_rv3.docx"),
    reviewer=reviewer,
)
print(out)  # manuscript_rv3.docx