feat: Add unrealized loss based early taker close

[keitaj]

Summary

• Add unrealized loss threshold (--unrealized-loss-close-bps) to PositionCloser: when a position's unrealized loss exceeds the configured bps, it is immediately closed via taker order regardless of position age
• Targets large-loss trades that accumulate while waiting for maker close or age-based force close
• Default 0 (disabled) preserves full backward compatibility

Changes

File	Change
strategies/mm_position_closer.py	Add CLOSE_REASON_UNREALIZED_LOSS constant, unrealized_loss_close_bps parameter, and unrealized loss check in manage() before age-based force close
strategies/market_making_strategy.py	Pass unrealized_loss_close_bps from config to PositionCloser
bot.py	Add --unrealized-loss-close-bps CLI argument and wire it into strategy params
README.md	Document the new parameter in strategy description and YAML reference
tests/test_unrealized_loss_close.py	9 test cases covering long/short, below/above threshold, profit, disabled, no market data, close reason logging, and existing close order cancellation

Test plan

• All 9 new tests pass
• Full test suite passes (753 tests)
• flake8 lint clean
• No regressions in existing tests

🤖 Generated with Claude Code

[keitaj]

Review: Unrealized loss early taker close

Clean, well-scoped feature with solid test coverage. The implementation follows existing patterns in PositionCloser correctly, backward compatibility is preserved via default=0, and the bps calculation is correct for both long and short positions.

---

1. Missing _consecutive_close_failures cleanup (nit)

strategies/mm_position_closer.py L253-256:

When the unrealized loss path closes a position, it pops _open_positions but does not clean up _consecutive_close_failures — unlike the force-close path (L342, L372) and cleanup_closed() (L161) which both do.

self._record_close(coin, CLOSE_REASON_UNREALIZED_LOSS, age, current_tier)
close_position_fn(coin)
self._open_positions.pop(coin, None)
return

The stale entry gets cleaned up on the next successful close placement for that coin, so this is not a functional bug — just an inconsistency with the other close paths.

Suggestion:

self._record_close(coin, CLOSE_REASON_UNREALIZED_LOSS, age, current_tier)
close_position_fn(coin)
self._open_positions.pop(coin, None)
self._consecutive_close_failures.pop(coin, None)
return

---

Verdict: LGTM

Single nit (non-blocking). Logic is correct, risk guards untouched, tests are thorough (9 cases covering all edge cases), and CLI/config wiring is complete. Ready to merge after CI passes.