fix: include retrieved memory content in Claude prompts

[cafitac]

Summary

• include short retrieved-memory snippets in codex-prompt and claude-prompt output instead of only metadata/policy lines
• fix Claude wrapper usability so the real Claude CLI can answer from retrieval-backed content in one-shot flows
• document the updated prompt/wrapper behavior and Claude readiness status

Test Plan

• TMPDIR=/Users/reddit/Project/agent-memory/.tmp-test uv run pytest tests/test_cli.py::test_python_module_cli_claude_prompt_outputs_plain_prompt_text tests/test_prompt_wrapper_scripts.py::test_run_claude_with_memory_dry_run_outputs_command_and_prompt -q
• TMPDIR=/Users/reddit/Project/agent-memory/.tmp-test uv run pytest tests/test_cli.py tests/test_prompt_wrapper_scripts.py -q
• python scripts/run_claude_with_memory.py /var/folders/y_/_1fc1fgn4kn5q1c6xccmjxv40000gn/T/agent-memory-claude-smoke-tfsany5m.db "What is the Claude Code wrapper smoke target? Reply with the target phrase only." --preferred-scope project:agent-memory --max-turns 3

Notes

• an unrelated repo-wide CLI test run initially hit OSError: [Errno 28] No space left on device under the default macOS temp directory; rerunning with TMPDIR=/Users/reddit/Project/agent-memory/.tmp-test passed
• leftover untracked .agent-learner/, .claude/, .dev/kb/retrieval-eval-m1-implementation-plan.md, and .omc/ are pre-existing local artifacts and are intentionally excluded from this PR