strict_shape_contract hashes comments: comment-only edits demand reviewed repins (3 today) #1962

tools/strict_shape_contract.py pins each exempt builtin by the SHA-256 of its raw body and of every helper in its closure. Comments are part of that text, so an edit that changes only a comment fails the gate ("exempt source/helper changed; review the contract") exactly as a code change does. Because tools/strict_differential.sh aborts its preflight on the same hashes, one such edit fails about seven full-lane jobs at once.

What it cost on 2026-10-24:

• R1.6z needed two reviewed repins (builtin_exec_capture, builtin_tensor_load) where only a constructor's spelling inside a comment had changed;
• runtime: one replay-refusal helper, and the taped-fs rule reads k_tape_kinds (#1943) #1960 failed seven full-lane jobs, and its reviewed repin was a one-line code change;
• 181 fs: fail-soft tags in src/ have had no reader since #1318, and they have drifted #1949's planned sweep of 179 comment tags would force a repin for every exempt body it touches.

Each review was a closure diff whose answer was "comments only", and a gate whose reds are mostly false invites rubber-stamp repins.

Proposal: hash a normalized body, with comments stripped by a real tokenizer (string, character-literal and comment state; mechanical-gates §57) and whitespace collapsed. The gate still reds on any token change in the function or its helpers.

Done when

• The body and helper hashes are taken over the comment-stripped, whitespace-normalized token stream. The stripper honours string and char literals, and its selftest covers /* inside a string and // inside a char literal.
• A comment-only edit to an exempt builtin passes. A one-token code change in the builtin, or in a helper of its closure, fails, naming the function. Both are planted in the selftest.
• All existing contracts are repinned once, in the same PR, with a closure diff showing that only the hash function changed.