[codex] Add configurable secondary agent runtime

[devon7y]

What changed

• introduced a centralized secondary_agent.py runtime profile for the repo's heavier coding flows
• switched task spawning, research execution, work mode, QA retries, and self-fix entry points to use the shared secondary-agent helpers
• updated JARVIS prompt text and voice-command handling to refer to a configurable coding agent instead of hardcoded Claude-specific wording
• added focused tests for secondary-agent command construction and updated the feedback-loop test to mock the shared helper

Why

The repo previously hardcoded Claude-specific commands and language across multiple execution paths. That made it harder to change the coding runtime coherently and duplicated command-building logic in several places.

Impact

• the coding-agent runtime is now configured in one place
• switching the repo's heavier work agent is substantially simpler and lower risk
• Terminal launch, build, research, QA, and work-mode flows stay aligned with the selected runtime

Validation

• pytest -q tests/test_secondary_agent.py tests/test_feedback_loop.py
• note: tests/test_feedback_loop.py still reports 3 skipped async tests in this environment because pytest-asyncio is not installed in the worktree interpreter