docs(omo-senpi): teach the eval SDK import for every ulw-loop surface

Every Native guidance surface now drives the loop from a JS eval cell through
the SDK published at OMO_AGENT_TOOLKIT_SDK_ROOT instead of the removed
omo_agent_toolkit tool or a CLI spawn: the ulw-loop and ulw-research skills,
the skill-pointer sentence, the steering reminder and continuation prompt, the
ulw-plan scaffold wording, and the omo-native `ulw-loop` refusal message.
The docs QA is inverted and renamed to agent-toolkit-eval-sdk-docs.mjs: it
requires the import line and forbids tool/CLI forms across sources and the
generated copies.

Plan: .omo/plans/omo-agent-toolkit-eval-sdk.md