ConversationLearner: add non-interactive CLI entrypoint

[weiyilong-1]

What

Adds a non-interactive CLI to agents/conversation_learner, alongside the existing adk run REPL and the Agent Engine deployment (deploy.py), so the agent can run from scripts / cron / CI.

Two input styles:

• Deterministic flags — --reasoning_engine_id / --conversation_id + a time window (--days_ago or --start_time/--end_time), plus --include_ids. Calls generate_learnings() directly (no LLM front-end), so it is fast and reproducible.
• --prompt — routes a natural-language request through the existing root_agent one-shot via ADK InMemoryRunner (mirrors adk run).

--output controls where proposal.json is written (a directory is allowed). It works in both modes via a module-level default, so generate_learnings()'s ADK tool signature is left unchanged.

Why

The agent could previously only be driven interactively (adk run .) or deployed to Agent Engine. There was no scriptable entrypoint for automation/pipelines.

Files

• cli.py (new) — argparse CLI: build_parser(), _run_flags(), _run_prompt(), main().
• __main__.py (new) — enables python -m conversation_learner.
• agent.py — set_default_output_path() + an output_path fallback in save_trajectory_analysis_result (~6 lines; generate_learnings untouched).
• tests/test_cli.py (new) + output-path coverage added to tests/test_agent.py.
• README.md — "Running via CLI (non-interactive)" section.

No new dependencies (argparse / asyncio are stdlib).

Usage

# Deterministic flags (from agents/)
python -m conversation_learner \
  --reasoning_engine_id projects/<project-number>/locations/us-central1/reasoningEngines/<engine-id> \
  --days_ago 20 --include_ids

# Natural language
python -m conversation_learner --prompt "generate learnings for <reasoning-engine> in the past 20 days"

Testing

• python -m unittest conversation_learner.tests.test_cli conversation_learner.tests.test_agent — 104 tests pass.
• Verified all three invocation paths (adk run, CLI + flags, CLI + --prompt) against a live reasoning engine: each exits 0 and writes a valid proposal.json.