feat: add deepwolf replay command

[MichisGitIsKing]

Hey @JuneQQQ!

Here's the replay command you asked for in #16.

What's new

deepwolf replay PATH — loads a saved transcript JSON and re-renders the game to the terminal, same coloured event stream as simulate.

• transcript.py: added Transcript class, load(), and _event_from_json() to reconstruct Event objects from saved transcripts
• cli.py: new replay subcommand with optional --delay flag (seconds between events, default 0 = instant)
• 4 new tests: load reconstruction, round-trip (simulate → save → load → verify), event type integrity, and CLI end-to-end

All three CI checks pass (ruff, mypy, pytest — 63/63).

The replay just re-emits stored events through the existing renderer — no GameState rebuild needed, exactly as you suggested.

Closes #16

[JuneQQQ]

Thanks for picking this up @MichisGitIsKing — and great follow-through on #16. 🙏

The implementation is clean: reusing _print_event for the event stream is exactly right (no GameState rebuild needed, just re-emit through the renderer), and the four tests are well-scoped. Two real issues to address before this can merge, plus one small nit:

1. The bilingual invariant is dropped on the result panel

deepwolf has a strict bilingual invariant (en/zh via deepwolf.i18n, see #15) — every player-facing string goes through Translator so the whole game (event log, roles, result panel) renders in the chosen language. Your replay rebuilds the result panel inline and hardcodes English:

• Transcript.winner_label returns "the Village" / "the Werewolves" (transcript.py)
• cmd_replay writes "survived" / "died day N" and the panel title as English literals (cli.py)

A Chinese-language transcript replayed today would render its event log in Chinese (event .text is stored verbatim) but its result panel in English — that's the regression.

The cleanest fix is to store lang in the transcript itself and use the existing Translator for the panel. Concretely:

# transcript.py to_json — add one line:
"lang": ...   # I can point you at where to pull this from GameResult

# Transcript / cmd_replay:
from deepwolf.i18n import Translator, pick
tr = Translator(self.lang)
title = pick(self.lang, f"{tr.faction_label(self.winner)} win",
                       f"{tr.faction_label(self.winner)}获胜")
# and use pick(..., "survived", "存活") / pick(..., f"died day {d}", f"第 {d} 天死亡")

For the lang field on GameResult — currently it isn't stored on the result, only on GameConfig. Easiest path: thread it through to_json by also accepting (or reading from) the config, or add lang to GameResult. I'm fine with either; flag if you'd like guidance.

2. The branch is dirty against current main

Calibration follow-ups landed since you forked (#21, #22), so cli.py has new neighbours where you're inserting cmd_replay and the replay subparser. A rebase should resolve it without code changes — both additions are non-overlapping with the calibration code.

3. Tiny nit

A negative --delay (user typo) would raise from time.sleep. A max(0.0, args.delay) or a parser-level type= validator would be friendly.

Otherwise

• _event_from_json correctly inverts event_to_json (frozenset reconstruction, EventType conversion). ✅
• The CLI end-to-end test (test_replay_cli_runs_without_error) is a nice touch.
• I've approved the workflow run so you can see the full CI go green on this branch.

Rebase + the bilingual fix and I'll merge promptly. Sorry for the moving target on cli.py — the calibration work landed in parallel.