refactor: keep comments focused on contracts and constraints

[Mathih13]

Long comments repeat code and retain stale explanations. Remove that narration while keeping short caller contracts, protocol facts, safety constraints, tool annotations and documentation pointers. Rust comment lines fall from 47,350 to 1,005. Longer verified reasoning stays in the project documents.

Humans and agents follow one comment policy. Agent handoffs, delegated work, PR filing and PR review now require a separate comment-review result. This is a review requirement, not a new CI gate.

What you verified

• Comment review passed across the cleanup and its repairs. Independent reviews checked retained facts and lost context; their findings were corrected. Two fresh-agent trials checked implementation and review behavior.
• 4,857 Rust tests passed during cleanup. The later comment repairs preserve executable tokens; 45 navigation tests also passed. The final rebase includes the Bun update and the movement fix from fix(movement): roots hold a Character and moving ends Shadowmeld #696. All eight Rust files in that movement fix retain the exact executable tokens and strings from main. Its receipt constraints now live in a linked document.
• Final rebase checks passed: the root and receipt tests, the shared hold test, all 1,581 Module unit tests, formatting and bindings freshness. Independent review passed for the conflict resolution and moved receipt constraints.
• After the Bun rebase: bun install --frozen-lockfile and bun test --timeout 30000 on Bun 1.4.3, with 76 tests passed.
• Formatting, workspace Clippy, the Module wasm check, bindings freshness, four Python tests and five offline shell suites passed.
• Rust token checks, script syntax-tree checks and parsed configuration comparisons checked behavior preservation. The deliberate exceptions are documentation lint settings, formatting, emitted comment text and the regenerated fixture source hash.

What you did not verify

Durable integration and client suites were not run locally. No live Realm was used.

Risk

This touches 715 files. Review the policy and moved contracts in the documents alongside the source deletions; passing tests cannot prove that every useful explanation survived.

Lead model: GPT-6-Astra. Subagent checks: GPT-6.1-Sol and GPT-6-Astra. Harness: Codex in T3 Code.