fix(shared): stop run_game masking crashes with sys.exit in finally

[dieterolson]

Problem

run_game() in src/games/shared/game_launcher.py called sys.exit() inside a finally block:

finally:
    pygame.quit()
    sys.exit()

sys.exit() raises SystemExit, which replaces any in-flight exception. Any real crash — asset load failure, display init failure, a bug in game.run() — was silently discarded and the process exited 0, reporting success.

Confirmed empirically on a Raspberry Pi running Batocera: Force Field failed to create its display and exited 0 with no traceback whatsoever. The underlying error (pygame.error: <driver> not available from pygame.display.set_mode in src/games/shared/fps_game_base.py:133) was only recoverable by constructing the Game class directly, bypassing run_game.

This affects every game routed through run_game() — Force Field, Duum, and Zombie Survival at minimum.

Fix

Log the exception with a traceback and re-raise it, and move sys.exit() out of finally so it only runs on the success path:

except Exception:
    logger.exception("Game crashed")
    raise
finally:
    pygame.quit()

# Only reached when the game exited normally (or was interrupted).
sys.exit()

sys.exit() is kept on the success path rather than dropped entirely, so normal-exit and KeyboardInterrupt behavior is unchanged — both pre-existing tests pass unmodified. The contract is documented in a Raises: block per the repo's DbC convention.

Tests

Three regression tests added (written failing first, per the repo's TDD convention):

• exception from game.run() propagates out of run_game()
• exception from game construction propagates — this is the actual Force Field case, where set_mode fails in __init__
• the crash is logged with exc_info before being re-raised

These deliberately do not patch sys.exit — patching it is precisely what hid this bug from the existing test suite.

Verification

• RED → GREEN on the three new tests; tests/shared/test_game_launcher.py 11/11 passing.
• Full suite: 1826 passed, 2 skipped (pre-existing skips).
• ruff, black, mypy clean; pre-commit run --files on both changed files: all hooks passed.
• End-to-end subprocess check reproducing the Batocera failure (pygame.error raised from __init__): now prints a full traceback and exits 1. Before this change: exit 0, silent.

[github-actions]

⚠️ SPEC.md Update Required

Source files were modified in this PR but SPEC.md was not updated.

Per the Specification-Driven Development Policy:

• Any PR that adds, removes, or changes functionality must update SPEC.md
• Update the relevant sections (features, architecture, tests, dependencies, etc.)
• Bump the Spec Version if making substantive changes

To resolve:

1. Update SPEC.md to reflect your changes, OR
2. Add the spec-exempt label if this PR genuinely doesn't affect the spec (e.g., pure refactor with no behavior change)

This check is enforced by the spec-check CI workflow.