fix(agent): restore modern domain context after restart

[dajiaohuang]

Problem

After a persisted Narrative conversation and Hosted Agent restart, the MCP reconnects but pre-turn authoritative context synchronization raises unsupported schema.

Modern MCP wrappers intentionally remove trusted campaign_id from the model-visible input schema and retain it in _trusted_arguments. The synchronizer only recognized a visible action + campaign_id pair, so the exact modern campaign_query shape was rejected before the wrapper could inject the Host-owned campaign.

Solution

• recognize the modern shape only when action remains visible and the MCP wrapper explicitly advertises trusted campaign_id
• pass only action=get for that shape, letting the bound request context inject/override campaign identity
• bind a missing request campaign from the persisted authoritative binding
• reject a conflicting request campaign before any tool executes
• keep legacy view + payload and visible action + campaign_id compatibility
• continue rejecting action-only or otherwise unknown schemas

Verification

• uv run ruff check nanobot --select F
• uv run ruff check nanobot/agent/loop.py tests/agent/test_domain_context.py
• uv run python -m pytest -q tests/agent/test_domain_context.py (16 passed)
• uv run python -m pytest -q tests/agent/test_domain_context.py tests/tools/test_mcp_tool.py tests/tools/test_mcp_v2_contract.py tests/apps/test_hosted_worker.py (141 passed)
• uv run python -m pytest -q tests (5393 passed, 73 skipped)

Regression coverage persists and reloads a Narrative session across a simulated process restart, registers a freshly reconnected modern campaign tool, verifies v2 trusted campaign injection, and covers unsupported untrusted schemas plus cross-campaign rejection.