Real-time sync workflow: instruction injection for agent memory

[thrr87]

Summary:

• add hoard instructions command with dry-run, confirmation, and target selection
• integrate instruction injection into hoard setup with --no-instructions and safe non-interactive behavior
• move instruction planning/upsert/root-detection logic into hoard/cli/instructions.py
• expand generated OpenClaw client to argparse subcommands and add memory_put, sync, and inbox_put
• add CLI/module tests for root detection, marker upsert behavior, dry-run/apply/idempotency, and setup integration

Validation:

• ./.venv/bin/ruff check hoard/cli/main.py hoard/cli/instructions.py tests/test_cli_instructions.py tests/test_cli_instructions_module.py
• ./.venv/bin/python -m pytest -q tests/test_cli_instructions_module.py tests/test_cli_instructions.py tests/test_cli_init_setup.py

Notes:

• docs updates are excluded from this PR and handled separately.