fix(cli): fully filter session_meta from restored conversations

[Disaster-Terminator]

What does this PR do?

This PR fixes the CLI restored-session path so transcript-only metadata such as session_meta does not leak back into provider-facing conversation history.

The issue is that resumed CLI conversations could restore raw session rows and reuse them as normal chat history. When those rows included non-transcript entries like session_meta, strict chat-completions providers could reject the outgoing request because of unsupported roles.

This change fixes that in two places:

1. CLI restored history is now normalized before being reused, so resumed conversations only keep transcript-safe roles.
2. The shared API-boundary sanitizer also drops unsupported roles before sending provider requests.

This keeps the fix narrow and defensive:

• restored CLI conversations behave consistently
• internal metadata rows do not re-enter conversation_history
• unsupported roles do not cross the provider boundary even if they appear upstream

Related Issue

Fixes #4715

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 🔒 Security fix
• 📝 Documentation update
• ✅ Tests (adding or improving test coverage)
• ♻️ Refactor (no behavior change)
• 🎯 New skill (bundled or hub)

Changes Made

• Added shared restored-conversation normalization in cli.py
• Routed all CLI restored-session entry points through the same normalization path:
  • _init_agent()
  • _preload_resumed_session()
  • _handle_resume_command()
• Ensured restored conversation history drops non-transcript roles such as session_meta
• Kept API-boundary filtering in run_agent.py so unsupported roles are removed before provider calls
• Added regression tests covering:
  • _sanitize_api_messages() dropping unsupported roles
  • _preload_resumed_session() filtering restored history
  • _init_agent() resumed-session loading behavior
  • _handle_resume_command() filtering /resume history

How to Test

1. Create or use a saved CLI session whose restored rows include session_meta
2. Resume the session through CLI startup resume or the /resume command
3. Confirm that restored conversation_history contains only transcript-safe roles
4. Confirm that provider-facing requests no longer include unsupported roles such as session_meta
5. Run:
   pytest -q tests/test_agent_guardrails.py tests/test_resume_display.py tests/test_cli_resume_command.py

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):, feat(scope):, etc.)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run pytest tests/ -q and all tests pass
• I've added tests for my changes (required for bug fixes, strongly encouraged for features)
• I've tested on my platform: Ubuntu 24.04.4 LTS (WSL2), Python 3.11.15

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) — or N/A
• I've updated cli-config.yaml.example if I added/changed config keys — or N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — or N/A
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide — or N/A
• I've updated tool descriptions/schemas if I changed tool behavior — or N/A

Screenshots / Logs

Targeted regression tests passed:

pytest -q tests/test_agent_guardrails.py tests/test_resume_display.py tests/test_cli_resume_command.py
60 passed in 8.05s

I also ran `pytest tests/ -q` for broader signal, but did not use it as a merge gate because the repository test suite surfaced unrelated distributed failures outside this change set before a clean final summary could be captured.