fix: extract evolved skill body from optimized predictor

[steezkelly]

Summary

• embeds the skill body in the DSPy predictor instructions instead of passing it as a normal input field
• extracts the evolved skill body from the optimized predictor's mutated instruction/docstring surface using a sentinel
• adds regression coverage proving the saved evolved body changes when the optimized predictor text changes

Scope control

This is the focused split from the broad PR #39 for the ghost-improvement bug only.
It intentionally does not include:

• provider/model routing changes
• GEPA metric/threaded LM changes
• generated output/ artifacts
• Kanban board-state churn

Test plan

• /home/steve/hermes0/hermes-agent-self-evolution/venv/bin/python -m pytest tests/skills/test_skill_module.py tests/skills/test_evolve_skill_extraction.py -q
• /home/steve/hermes0/hermes-agent-self-evolution/venv/bin/python -m pytest -q

Result: 144 passed, 11 warnings.

Issue #38 relationship

This PR addresses the ghost-improvement extraction portion of #38: the optimized predictor must expose the actual evolved skill body, rather than reporting improved scores while saving the baseline body.

It intentionally leaves the remaining #38 follow-ups open for focused PRs:

• provider/model routing and Nous API integration changes
• GEPA API/threaded LM compatibility changes
• constraint validator body-substance checks
• JSON robustness fixes
• generated output artifact cleanup

[steezkelly]

Local verification refresh from a clean temporary checkout of this PR:

• gh pr checkout 49 --repo NousResearch/hermes-agent-self-evolution
• /home/steve/hermes0/hermes-agent-self-evolution/venv/bin/python -m pytest tests/skills/test_skill_module.py tests/skills/test_evolve_skill_extraction.py -q → 12 passed, 11 warnings
• /home/steve/hermes0/hermes-agent-self-evolution/venv/bin/python -m pytest -q → 144 passed, 11 warnings

GitHub currently reports no visible Actions runs/checks for this branch, so local verification is the available signal for now.

[steezkelly]

Update: added a minimal GitHub Actions pytest workflow to this branch so the repo has an in-repo CI definition going forward.

Local verification after the workflow commit (0fe6673421b201e63b0247ef11658b4c6e933313):

cd /tmp/self-evolution-pr49-ci
/home/steve/hermes0/hermes-agent-self-evolution/venv/bin/python -m py_compile $(find evolution -name '*.py' -print)
/home/steve/hermes0/hermes-agent-self-evolution/venv/bin/python -m pytest tests/skills/test_skill_module.py tests/skills/test_evolve_skill_extraction.py -q
# 12 passed, 11 warnings
/home/steve/hermes0/hermes-agent-self-evolution/venv/bin/python -m pytest -q
# 144 passed, 11 warnings

Note: GitHub currently shows no regular Actions runs for this repo/branch; upstream only has the Copilot code review workflow visible. The new workflow is included in the PR so future pushes/PRs can run python -m pytest -q once Actions picks it up/after merge.