Add context contract tests

## Problem
The two Skills share Review Target / Review Context / Existing Review Evidence semantics, but there is no cross-Skill contract test proving their shared fields stay aligned while mode-specific differences remain intentional.

## Goal
Add focused input-contract tests covering representative inputs for both Skills.

## Scope
- Representative Target / Context / Existing Review Evidence inputs for both Skills.
- Shared-field parity assertions across the pair.
- Assertions that intentional local-vs-GitHub differences are preserved, not collapsed.
- Invalid-input rejection for both Skills.
- Runs in standard validation (discovered by the existing test suite).

## Non-goals
- No output-schema work (see #44/#67-#71).
- No redesign of the shared contracts in `shared/policies/review-context.md` / `review-evidence.md`.
- No new review semantics.

## Dependencies
Depends on #72's residual local Review Target contract work landing first (there is currently no structured local-mode target to test against).

## Validation
- Test run.

Read shared/policies/review-context.md and shared/policies/review-evidence.md, then inspect both Skills and the residual local Review Target work from #72. Run the standard test suite first. Done means representative valid and invalid inputs are covered, shared fields stay in parity, and intentional local-versus-GitHub differences remain asserted.

Written by the indexing model from the issue text.