Draft spec_commit is parsed but never compared: a draft from another Spec revision is resumed and published #223

This blocker comes from the external review of #221 (ChatGPT batch 30, claim B7). It also implements the owner's decision of 8 November 2026: a draft saved against a different pinned Spec commit is refused at resume, and a fresh draft is derived from the folders.

draft_format.answers_from_bytes() accepts any 40-digit lowercase hex spec_commit (draft_format.py:218-226). run/draft_file._resume() checks only base_sha256 and then applies the stored answers (run/draft_file.py:127-136). publish_draft() then treats those answers, including repeat confirmations and ID choices, as current. A draft written under another Spec revision is therefore published without the user re-confirming anything under the current rules.

Fix:

• In _resume(), and in save_draft() after the strict round trip, require answers.spec_commit == SPEC_COMMIT.
• Otherwise raise DraftFileError with a dedicated code such as spec_commit_mismatch. The message tells the user that the published file is untouched and that a fresh draft will be derived from the folders.
• Never rewrite the stored commit.

Regression test: start from a complete valid draft and change only spec_commit to another valid 40-hex SHA. Then check that:

• load_draft refuses;
• publish_draft refuses;
• morphobricks.toml is unchanged;
• the draft bytes are byte-identical;
• no confirmation is carried forward.