fix: agentkit test hygiene — dead branch, misleading name, fragmented drift detectio

[yvonnedevlinrh]

Follow-up from PR #59 review. Four non-blocking test quality findings in internal/agentkit/agentkit_test.go:

F1/F10 — Dead negative test in TestForgeGlobal_StructuralHardening

Assertion (6) checks for a line starting with "3." that also contains "ttl_seconds" and "auto-release". The current step 3 is "3. FINALLY, coordinator can emergency release..." which doesn't contain "ttl_seconds", so the branch can never fire. The regression it guards against (standalone TTL step reintroduced) would not be caught.

Suggested fix: match any line containing both "ttl_seconds" and "auto-release" that is NOT part of the comms_reserve call, regardless of step number.

F4 — TestSkillFiles_DriftDetection name implies generic scope

The name says "SkillFiles" but the implementation covers only always-on-guidance and forge-global (2 of 7 skills). Either scope the name (TestHardenedSkillFiles_DriftDetection) or extend to all 7 skills.

F2 — Fragmented drift detection

Drift detection is split across two tests covering disjoint subsets: TestSkillFiles_DriftDetection (2 skills) and TestForgeCoordinationSkill_StructuralHardening assertion (5) (1 skill). Four skills have no drift detection. Consider consolidating into a single parameterized test covering all 7.

F3 — t.Fatalf in loop masks failures

TestSkillFiles_DriftDetection uses t.Fatalf for ReadFile errors inside the skill loop. If both skills fail to read, only the first is reported. Use t.Errorf + continue instead.