feat: add strong-exploration campaign mode

[ccomkhj]

Summary

• Add strong-exploration campaign mode with relaxed guardrails for aggressive parameter exploration
• In this mode: revise_search allowed every round, no churn limit, no cooldown between narrow/widen
• Exploration reset: when patience triggers in strong-exploration mode, the system auto-selects a new param subset (coverage-based, prioritizing untried params) and keeps going instead of terminating
• Only hard stops (max_wall_time, max_rounds, target_score) terminate the campaign — patience just triggers a reset
• Mode stored in CampaignConfig, persisted in DB, exposed via CLI (--mode) and MCP

Key Changes

• src/agentune/core/models.py — mode field on CampaignConfig
• src/agentune/core/db.py — mode column on campaigns, reset_number column on study_rounds
• src/agentune/core/campaign.py — mode-aware guardrail bypass (eligibility, churn, cooldown)
• src/agentune/runner.py — patience scoped by reset_number; auto-reset in strong-exploration
• src/agentune/exploration.py — coverage-based param selection for resets
• src/agentune/cli.py — --mode flag

Test Plan

• 248 tests passing, 0 failures
• Unit tests for param selection coverage logic
• Guardrail bypass tests (eligibility, churn, cooldown)
• Auto-reset integration tests (patience triggers reset, not completion)
• Standard mode regression (patience still terminates)
• CLI smoke test with exploration reset

🤖 Generated with Claude Code