Bug: _check_skill_structure validates `body` (no frontmatter) instead of full skill text, causing false failures

In evolution/skills/evolve_skill.py, inspect the baseline and evolved validate_all call sites around lines 122 and 189, then compare them with skill_module.py's raw/body definitions. Reproduce with --eval-source sessiondb or synthetic and confirm valid skills no longer produce evolved_FAILED.md and the constraint log passes skill_structure.

Summary

ConstraintValidator._check_skill_structure() always receives skill text without YAML frontmatter, so it perpetually fails with "Skill missing: YAML frontmatter (---)" even when the skill file is perfectly valid.

Root Cause

In evolution/skills/evolve_skill.py, two call sites pass the wrong slice of the parsed skill dict to validate_all():

Baseline check (line ~122):

# BUG: skill["body"] is the markdown body only — no frontmatter
baseline_constraints = validator.validate_all(skill["body"], "skill")

Evolved skill check (line ~189):

# BUG: evolved_body is also frontmatter-stripped; evolved_full has it
evolved_constraints = validator.validate_all(evolved_body, "skill", baseline_text=skill["body"])

skill_module.py documents the dict clearly:

• "raw" — full file content (frontmatter + body)
• "frontmatter" — YAML between --- markers
• "body" — markdown after frontmatter

_check_skill_structure looks for text.strip().startswith("---"), which can never match when called with "body".

Impact

• Every evolved skill is marked FAILED regardless of actual content quality
• Evolved output is saved as evolved_FAILED.md and never deployed
• The constraint gate is entirely non-functional for the skill_structure check
• baseline_text passed to _check_growth is also wrong ("body" vs "raw"), making the growth-limit comparison slightly incorrect

Fix

# Line ~122 — baseline check
baseline_constraints = validator.validate_all(skill["raw"], "skill")

# Line ~189 — evolved check  
evolved_constraints = validator.validate_all(evolved_full, "skill", baseline_text=skill["raw"])

evolved_full is already computed just above via reassemble_skill(skill["frontmatter"], evolved_body).

Steps to Reproduce

1. Run any skill evolution with --eval-source sessiondb or synthetic
2. Observe that evolved output always lands in output/<skill>/evolved_FAILED.md
3. Check the constraint log — skill_structure will always show ✗

Environment

• hermes-agent-self-evolution cloned from main (tested 2026-05-10)
• DSPy fallback to MIPROv2 (GEPA unavailable in installed version)
• Both video-transcoding and audio-transcription skills affected