refactor(libsy): extract load_judge_config and drop reasoning fallback in judge verdict parsing

[linj-glitch]

Two related judge changes: extracts the prompt+schema loading logic into a shared load_judge_config helper (used by both the capability classifier and the escalation judge), and drops the reasoning-content fallback in parse_json_verdict. The fallback was added to handle gateways that return verdicts in reasoning blocks, but the judge is now configured to run without thinking, so the completion always carries the verdict — reading from reasoning was masking truncated replies instead of surfacing them as errors. Part 3 of breaking up PR #172.

Summary by CodeRabbit

• Bug Fixes
  • Improved task-routing judge requests by consistently including required system guidance.
  • Fixed verdict parsing to read responses from the correct content and avoid treating reasoning text as a verdict.
  • Improved handling of structured response schemas used by judge evaluations.
• Tests
  • Added coverage for verdicts containing reasoning-only content and updated validation of judge request formatting.

[coderabbitai]

Walkthrough

Changes

Judge evaluation flow

Layer / File(s)	Summary
Shared judge configuration loader crates/libsy/src/algorithms/util/llm_judge.rs, crates/libsy/src/algorithms/util.rs	Adds shared schema parsing and prompt rendering, then re-exports the loader within the crate.
System message request construction crates/libsy/src/algorithms/llm_class.rs	Places the judge prompt at the start of LlmRequest.messages and updates the corresponding test expectations.
Completion verdict parsing crates/libsy/src/algorithms/util/llm_judge.rs	Parses verdicts from completion content, ignores reasoning blocks, and adjusts classification extraction.
Estimated code review effort: 3 (Moderate)	~20 minutes

Poem

I’m a rabbit with prompts in my pack,
A system message leads the stack.
Schemas bloom bright, verdicts come through,
Reasoning hops aside from view.
Tests thump softly: the path is true.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately reflects the main refactor and the verdict-parsing change in this PR.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

🧹 Nitpick comments (1)

crates/libsy/src/algorithms/llm_class.rs (1)

850-850: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Assert the prepended system-message contract.

The count alone passes if the first message has the wrong role or prompt. Assert that message zero is Role::System and equals judge.config.system_prompt.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@crates/libsy/src/algorithms/llm_class.rs` at line 850, Extend the assertions
around judge_request.llm_request.messages to validate the prepended message
itself: assert message zero has Role::System and its prompt/content equals
judge.config.system_prompt, while retaining the existing length assertion.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@crates/libsy/src/algorithms/llm_class.rs`:
- Line 850: Extend the assertions around judge_request.llm_request.messages to
validate the prepended message itself: assert message zero has Role::System and
its prompt/content equals judge.config.system_prompt, while retaining the
existing length assertion.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: c7b0fa6b-c8eb-46ca-a043-ba9a8640c536

📥 Commits

Reviewing files that changed from the base of the PR and between f3ff7ff and 6577642.

📒 Files selected for processing (3)

• crates/libsy/src/algorithms/llm_class.rs
• crates/libsy/src/algorithms/util.rs
• crates/libsy/src/algorithms/util/llm_judge.rs