feat: add JSONL event streaming, Textual replay UI, and Ollama support for Kahn simulations

[UnbornAztecKing]

Summary

• Add a new sim.py execution path that emits structured JSONL events alongside CSV output.
• Introduce a wargame_tui terminal UI for replaying runs, inspecting turns/events, and live-following simulations.
• Add a typed event pipeline (GameEvent, writer, store, tailer) to separate simulation output from presentation.
• Extend legacy runners with ollama: model support and minor response-parsing hardening.
• Add packaging, CI, fixtures, docs, and test coverage for the new workflow.

Details

• Simulation

  • Add sim.py with provider routing for OpenAI, Anthropic, Gemini, and ollama: models.
  • Emit structured events for lifecycle, phase transitions, LLM decisions, state changes, KPI updates, and reports.
  • Add --events-file to write JSONL to a file or stdout.
  • Use json-repair to tolerate malformed model JSON.

• TUI / events

  • Add typed event models with sequence numbers, timestamps, payloads, and content hashes.
  • Add append-only writer, JSONL-backed store, and polling tailer for replay/live-tail workflows.
  • Add a Textual UI with turn navigation, event cards, KPI display, and search.
  • Add sample fixtures and demo event data.

• Legacy runners

  • Update Kahn_game_v11.py and Kahn_game_v12.py to support ollama: model IDs via configurable OLLAMA_BASE_URL.
  • Harden parsing for empty or missing model responses.

• Tooling

  • Add pyproject.toml, uv.lock, .python-version, CI, Make targets, and README updates.

Risk & Impact

• Moderate risk: large additive change with a parallel sim.py path that may drift from legacy runners.
• Performance trade-off: event writes flush/fsync on append, and the TUI polls for updates.
• Current event storage is in-memory and better suited to smaller runs than large archives.
• Raises the tooling/runtime baseline with Python 3.12, uv, and new dependencies.
• Backward-compatibility risk is otherwise low; existing runner flows remain in place.

Testing

• Add unit tests for event models, JSONL storage, writer behavior, and serialization.
• Add Textual smoke tests for mount/render, turn selection, KPI updates, and search interactions.
• No obvious end-to-end coverage for sim.py parity with legacy runners or live-tail behavior.

Reviewer Notes

• Focus review on behavioral drift between sim.py and the legacy runners.
• --compact appears to be parsed by the TUI entrypoint but not clearly wired through.
• Manual QA:
  • Run sim.py --events-file ... and verify TUI live-follow behavior.
  • Validate event ordering, sequencing, and KPI updates during a live run.
  • Verify ollama: routing with custom OLLAMA_BASE_URL.