Clarify PR-B cache and measurement telemetry

[ictechgy]

Summary

• Bump audit feasibility schema to contextguard.metric-feasibility.v1.1 for additive cache-friendliness fields.
• Emit explicit cache-friendliness overlap telemetry and partial confidence when prefix/tail windows overlap or evidence is partial.
• Memoize build_cache_friendliness through shared audit output paths to avoid redundant recomputation.
• Add regressions for cache-discount recommendations staying separate from token reduction.
• Clarify benchmark/provider-cache caveats and example report shape-only guidance.

Verification

• python3 -m unittest discover -s tests -p 'test_context_guard_kit.py' -k cache_friendliness
• python3 -m unittest discover -s tests -p 'test_context_guard_kit.py' -k transcript_audit
• python3 -m unittest discover -s tests -p 'test_context_guard_kit.py' -k benchmark
• python3 -m unittest discover -s tests -p 'test_context_guard_kit.py' -k benchmark_report
• python3 -m unittest discover -s tests -p 'test_context_guard_kit.py' -k benchmark_runner_rejects_incompatible_existing_csv_schema
• python3 -m py_compile context-guard-kit/claude_transcript_cost_audit.py plugins/context-guard/bin/context-guard-audit context-guard-kit/benchmark_runner.py plugins/context-guard/bin/context-guard-bench tests/test_context_guard_kit.py
• cmp -s context-guard-kit/claude_transcript_cost_audit.py plugins/context-guard/bin/context-guard-audit
• cmp -s context-guard-kit/benchmark_runner.py plugins/context-guard/bin/context-guard-bench
• python3 -m json.tool docs/benchmark-report.example.json >/dev/null
• git diff --check
• python3 scripts/release_smoke.py
• python3 scripts/prepublish_check.py

Notes

• G010 PR-B only: no artifact receipt, diet scanner, or context packer follow-ups included.
• OMX team startup was attempted first for G002 but produced no worker completion evidence; fallback was recorded in the ultragoal ledger and implementation was completed with native sidecar review.