feat(envs): Add Sally Soprano negotiation environment - #167

Ready to merge. The blocker was that SallySoprano-v0 was never added to textarena/envs/__init__.py, so ta.make("SallySoprano-v0") raised KeyError. I synced the branch with main (conflict-free — it only adds new files) and registered it:

register_with_versions(id="SallySoprano-v0", entry_point="textarena.envs.SallySoprano.env:SallySopranoEnv", wrappers={"default": [LLMObservationWrapper], "-train": CONVERSATIONAL_WRAPPERS}, max_rounds=60)

Verified locally on 3.11: make → reset (3 players) → step all succeed; the 3-player requirement in reset is respected. Env code itself was left untouched. Good to go.

Thanks for this, @cstorm125. Rebased onto current main and opened #202. I also wired error_allowance through to the state (it was declared/tested but never passed), made repeated invalid moves forfeit to the opponent (they previously left the offender stuck on-turn, which could hang the game), and dropped a couple of dead reward writes. Your 18 tests pass plus deterministic checks for the no-deal draw, forfeit, and all judge verdicts. The live deal→judge path still needs a model for the judge — flagged that in the PR. Credit to you for the design; happy to land it there or here.