fix: update GEPA construction for DSPy 3

[steezkelly]

Summary

Fixes DSPy 3.x GEPA construction for skill evolution:

• removes the obsolete max_steps argument from dspy.GEPA(...)
• adds a reflection LM from the existing optimizer_model parameter
• uses bounded max_metric_calls as the DSPy 3.x optimization budget
• isolates GEPA construction in _create_gepa_optimizer(...) so future DSPy API drift is easier to test
• adds regression tests that verify max_steps is not passed, reflection_lm is set, and zero iterations clamp to at least one metric call

Root cause

DSPy >=3.1 removed max_steps from GEPA and requires a reflection LM or instruction proposer. The previous code always raised during GEPA construction and silently fell back to MIPROv2.

Test Plan

• pytest tests/skills/test_evolve_skill_gepa.py -q
• pytest -q
• runtime signature probe: inspect.signature(dspy.GEPA) confirms max_steps is absent and max_metric_calls / reflection_lm are present
• static added-line security scan for secrets/shell injection/eval/pickle/SQL string formatting
• independent code review pass

Result: 141 passed, 11 warnings (DSPy deprecation warnings only).

Closes #10

[steezkelly]

Closing this PR in favor of consolidated PR #68. Local integration found real helper-block overlap in evolution/skills/evolve_skill.py across the stack, and #68 preserves local test evidence: targeted stack tests 41 passed; full suite 164 passed; GitHub checks were absent on the split PRs. Review #68 instead.