Compress AI system prompt to core protocol contract

[GVCUTV]

Motivation

• Reduce token usage and remove duplicated/verbose normative prose and examples while preserving the strict AI output contract used by the system.
• Ensure the system prompt contains only the required schema contract, hard/soft precedence, role enum constraints, and output JSON format so runtime parsers remain deterministic.

Description

• Rewrote src/main/resources/ai/system_prompt_template.txt to a compact, bullet-based prompt preserving behavior and adding a short top-level "Non-negotiable rules" section.
• Retained the single-output variants JSON contract and the required variant labels EMPATHETIC, EFFICIENT, BALANCED, plus the schema for status, metadata, assignments, uncovered_shifts, and uffa_delta.
• Kept and emphasized hard-vs-soft constraint precedence and role constraints (only STRUCTURED, SPECIALIST_JUNIOR, SPECIALIST_SENIOR) and explicit prohibition of echoing internal artifacts like role_validation_scratchpad.
• Only changed src/main/resources/ai/system_prompt_template.txt and committed the change as "Compress AI system prompt to core protocol contract".

Testing

• Reviewed relevant parser/validator classes under org.cswteams.ms3.ai.protocol (including AiScheduleJsonParser, AiScheduleVariantsResponseDto, AiScheduleResponseDto, AiScheduleSemanticValidator, AiRoleCoveredDeserializer, and AiUffaQueue) to confirm the compressed prompt aligns with existing expectations, and those inspections passed.
• Attempted to run protocol tests with mvn -q -Dtest='org.cswteams.ms3.ai.protocol.*Test' test, but the build could not complete due to external Maven parent POM resolution failing with 403 Forbidden in this environment, so automated test execution did not finish.

---

Codex Task