fix(security): accept fully covered documentation scans

[deepujain]

Summary

• accept SkillSpector partial completeness only when the scan has 100% coverage and every analyzer is either completed or not applicable
• restrict the exception to reports that declare no executable scripts
• preserve fail-closed behavior for limitations, ledger exceptions, partial or uninspected files, executable skills, unavailable analyzers, and missing evidence
• add positive documentation-only coverage plus adversarial regression cases

Closes #137.

Verification

• .venv/bin/python -m pytest -q tests/validators/test_security.py — 342 passed
• make PYTHON=.venv/bin/python lint — passed
• .venv/bin/python -m build — sdist and wheel built successfully
• git diff --check origin/main...HEAD — passed
• make PYTHON=.venv/bin/python test — 6,061 passed, 18 skipped, 4 deselected, 4 unrelated baseline failures: one CLI golden-surface drift and three Anthropic SDK httpx2 compatibility expectations

Release Impact

Documentation-only skills can pass security validation when SkillSpector inspected the complete component inventory and reports code-dependent analyzers as not applicable. Reports with any incomplete evidence continue to fail closed.

[rng1995]

[P1] Verify which analyzers are not applicable. Reducing all evidence to a set of states lets any required zero-work analyzer activate this exception. Starting from the new docs-only payload, change mcp_tool_poisoning to status="not_applicable" with planned_work=completed=0; validation still passes, although SkillSpector 2.11.2 normally runs that analyzer over SKILL.md with planned/completed work of 1. This can turn missing MCP security analysis plus an explicit partial verdict into a clean scan. Match the exact allowed docs-only N/A analyzer IDs (and reason codes/version), and require every other required analyzer to be completed.