feat(#131): checkpointed pipeline + fix import path issues

[redbbean]

Description

Closes #131

Implements a checkpointed LLM extraction pipeline so that an interrupted run (container crash, Ollama timeout, Ctrl+C) resumes exactly where it left off without re-processing already-extracted fields.

Also fixes a ModuleNotFoundError: No module named 'src' that prevented make exec from running, caused by an incomplete PYTHONPATH. Fixes #116 and #118.

src/llm.py

• Session ID computed lazily at start of main_loop()
• Session ID hashes transcript and field names, so two different form templates on the same transcript won't share a checkpoint (addition to PR #131- stateful saving #133, which only hashes the transcript)
• _get_field_names() handles _target_fields as either a dict (from pypdf) or a list
• Atomic checkpoint writes via .tmp + os.replace()
• SIGINT handler so Ctrl+C flushes the checkpoint before exit
• Retry loop (3 attempts, backoff) on Ollama timeouts with checkpoint saved
• JSONL error log for -1 responses to help with debugging
• Renamed json=None param to json_data=None

src/main.py

• Added missing from typing import Union (pre-existing bug from issue #

docker-compose.yml + Dockerfile

• PYTHONPATH=/app/src changed to PYTHONPATH=/app:/app/src

docs/session_resumption.md (new)

• Documents checkpoint mechanism, env vars, and usage examples as required by [FEAT]: Stateful Resumption & Checkpointed Pipeline for LLM Extraction #131 acceptance criteria

Type of change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• This change requires a documentation update

How Has This Been Tested?

• PYTHONPATH fix: make exec runs successfully

• Checkpointing logic: has not been tested

• Test A — make exec completes without import errors

Test Configuration: Docker Desktop, python:3.11-slim, Ollama + Mistral

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• Any dependent changes have been merged and published in downstream modules

Tests for the checkpoint logic are in progress.

Coming across this issue was interesting because I've implemented a similar checkpointing and resuming process in LLM benchmarking, so I was able to reuse some of the error log, checkpoint write, and retry loop logic. Please let me know if there's anything I can improve!