Add configurable runtime cap support to comparison drivers and solvers

[sidhsikka1]

Motivation

• Provide a single, easily-configurable total runtime cap for method comparisons so each solver can be limited consistently during evaluations.
• Implement different timeout behaviors per solver: stop IPOPT/full NLP via a callback, stop spawning GA generations when time elapses, and for the decentralized island GA convert remaining migration budget into a time-limited migration phase that transitions to consensus when expired.

Description

• Added scripts/run_comparison.py with a TOTAL_RUNTIME_CAP_S constant and wired it into run_method_comparison via a new max_runtime_s parameter.
• Extended run_method_comparison to accept max_runtime_s and pass it to all three solvers (centralized_nlp, centralized_ga, decentralized_island_ga).
• Plumbed max_runtime_s into og_opts.full_nlp and added a wall-clock timeout callback to stop the trust-constr solver when the runtime cap is reached (spacecraft_libraries/og_opts.py and spacecraft_libraries/solvers/centralized_nlp.py).
• Modified centralized GA to run one generation at a time and check elapsed wall time between generations to stop early (spacecraft_libraries/solvers/centralized_ga.py).
• Updated decentralized island GA and GraphManager.run_island_evolution to accept max_runtime_s, scale the effective migration time by num_agents (parallelism), stop migration when the cap is hit, and then proceed to the consensus phase (spacecraft_libraries/solvers/decentralized_island_ga.py and spacecraft_libraries/graph/graph_manager.py).

Testing

• Compiled all modified modules with python -m compileall scripts/run_comparison.py spacecraft_libraries/evaluation/comparison.py spacecraft_libraries/solvers/centralized_nlp.py spacecraft_libraries/solvers/centralized_ga.py spacecraft_libraries/solvers/decentralized_island_ga.py spacecraft_libraries/graph/graph_manager.py spacecraft_libraries/og_opts.py, which succeeded.
• Performed a quick smoke attempt calling run_method_comparison(..., max_runtime_s=0.01), which could not complete due to missing runtime dependencies in the environment (ModuleNotFoundError: No module named 'numpy').

---

Codex Task

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 521294a82e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Stop multiplying runtime cap by number of agents

run_island_evolution measures elapsed wall-clock time, but it scales the limit as max_runtime_s * self.num_agents even though each generation is executed serially in the same process (for agent in self.agents). This lets the decentralized solver run up to N times longer than the configured cap (e.g., ~90s with 3 agents and a 30s cap), which breaks the intended per-method runtime limit and can skew comparison results.

Useful? React with 👍 / 👎.