fix: escalate Claude companion network calls

[behruznassre]

Summary

• require targeted require_escalated sandbox permissions whenever Codex launches the Claude companion
• cover foreground and background review/adversarial-review paths plus rescue forwarding
• add contract and Codex E2E assertions so these launch instructions cannot silently regress

Root cause

Codex's default workspace-write sandbox does not provide network access. The built-in forwarding paths launched claude-companion.mjs without requesting targeted escalation, so the Claude CLI could start but DNS resolution for the Claude API failed with ENOTFOUND.

The fix escalates only the companion shell call, with a narrowly scoped justification. It does not enable network access globally; Claude's own task/review sandbox and permission controls continue to govern repository access and mutations.

Validation

• test-first contract check failed in the three affected skill/runtime contracts before the implementation change
• node --test tests/skills-contracts.test.mjs: 9 passed, 0 failed
• real companion API smoke test under targeted escalation returned exactly CC_NETWORK_OK
• npm run check:
  • unit: 505 passed, 0 failed
  • integration: 41 passed, 0 failed
  • Codex E2E: 22 passed, 0 failed
• git diff --check: clean

Closes #81

[upwind-code-us]

Upwind IaC Scan - 🔍 Scan in progress…

Upwind is scanning this PR. Results will appear here when the scan completes.

[upwind-code-us]

Upwind Code Scan - 🔍 Scan in progress…

Upwind is scanning this PR. Results will appear here when the scan completes.

[sf-jin-ku]

Reproduced the boundary: Codex's default workspace-write sandbox has no outbound network, so the companion starts Claude and then fails DNS. Scoping the escalation to the single companion exec_command (rather than telling users to enable global network_access) is the right narrow fix, and the contract + E2E assertions keep the forwarding children from regressing to use_default.