feat: infer goal turn budgets up to 250

[unreliable-machine]

Summary\n- Treat goals.max_turns as a ceiling capped at 250 instead of a fixed budget for every /goal.\n- Infer per-goal budgets from prompt complexity: simple goals keep 20 turns; complex plan-style or explicit persistence prompts scale up automatically.\n- Update CLI/gateway config handling, docs, and focused goal tests.\n\n## Tests\n- python -m pytest -o 'addopts=' tests/hermes_cli/test_goals.py tests/gateway/test_goal_max_turns_config.py tests/tui_gateway/test_goal_command.py -q\n- python -m py_compile hermes_cli/goals.py cli.py gateway/run.py hermes_cli/config.py\n- Manual heuristic assertions for simple, complex, and explicit-turn prompts\n

[teknium1]

Thanks for the contribution. The premise is still live: current main assigns each goal the configured fixed budget in hermes_cli/goals.py:1143-1152.

Problems

• The PR updates goal-specific documentation but leaves the canonical configuration reference at website/docs/user-guide/configuration.md:842-851 describing goals.max_turns as a fixed 20-turn safety cap.
• Current defaults explicitly define that cap as protection against judge false negatives and unbounded spend (hermes_cli/config.py:2271-2284). Automatic escalation to 250 changes that contract, so the intended default/opt-in policy needs maintainer confirmation.

Suggested changes

• If approved, update the configuration reference and add CLI, gateway, and TUI coverage for simple versus escalated goals under a configured 250-turn cap, including a lower configured cap.

Automated hermes-sweeper review.