Normalize the Review Context input model

## Type

Refactor

## Area

Platform Contracts

## Priority

P2 — Medium

## Problem

A shared Review Target / Review Context / Existing Review Evidence model now
exists (`shared/policies/review-context.md`, `shared/policies/review-evidence.md`)
and both Skills apply it through thin, tested policy layers (#75, closed). What
remains is structural, not conceptual: the local Skill's Review Target and the
GitHub Skill's Review Context and Existing Review Evidence don't yet have the
same code-level structured/tested representation their counterparts have, and
there's no cross-Skill contract test proving shared fields stay aligned.

## Goal

Both Skills consume equivalent structured representations of Review Target,
Review Context, and Existing Review Evidence, proven by a cross-Skill contract
test suite.

## Scope

Child issues:

- [x] #75 — Align both Skills to the shared context model (documentation/policy layer — done)
- [ ] #72 — Local-mode Review Target structured contract
- [ ] #73 — GitHub-mode Review Context structured contract
- [ ] #74 — Local-mode Existing Review Evidence reviewed-state identity
- [ ] #76 — Cross-Skill context contract tests

## Non-Goals

- Does not erase local vs GitHub PR differences.
- Does not change review reasoning.
- Not the output contract (see #44).

## Acceptance Criteria

- [ ] #72, #73, #74, #76 are closed.
- [ ] Each child links back to this issue.

## Dependencies

Depends on: none
Related: #44 (independent output-contract work; no ordering dependency
between the two — see #44 for the corrected dependency picture)
Parent: #39

## Validation

- Context contract tests (#76) validate both Skills once #72–#74 land.
- Both Skill validation suites pass; doc links resolve.