scaffold: per-tool rule files diverge in mandatory-workflow ordering

[vivek7405]

Problem

Surfaced by the self-review loop on PR #133. The scaffold ships three per-tool rule files that are supposed to "say the same thing" per the AI-first-is-plumbing blog post (blog/ai-first-is-plumbing.md:29). They diverge structurally:

• packages/cli/templates/.cursorrules orders the mandatory list as TESTS (1), DOCS (2), CONVENTIONS (3), PRE-MERGE SELF-REVIEW LOOP (4). Tests collapses server + browser into one bullet.
• packages/cli/templates/.agents/rules/workflow.md (the new Antigravity file from scaffold: rename Windsurf surfaces to Antigravity (post-merger) #132) orders the same list as Server tests (1), Browser tests (2), Documentation (3), Convention check (4), Pre-merge self-review loop (5). Tests are split.
• packages/cli/templates/.github/copilot-instructions.md uses a six-item list with "commit per logical unit" as item 1; the other two files put commit-per-unit under a separate Git rules section.

Not a substance bug. Same rules in all three. But an AI agent reading two files for the same project sees three different shapes, and a future edit that updates one file will probably miss the others.

Design / approach

Pick the canonical shape. Strong candidates:

1. Five-item list shape (Server tests, Browser tests, Documentation, Convention check, Pre-merge self-review loop) per workflow.md. Reads cleanly because each item is a single concern.
2. Four-item list shape (Tests, Docs, Conventions, Pre-merge self-review loop) per .cursorrules. Reads more compactly.

Pick one, apply to all three files. The commit-per-logical-unit rule stays under Git rules in all three to match where it lives today.

Acceptance criteria

• Pick the canonical shape (note the decision in the PR body).
• All three files share the same numbered list under "Mandatory workflow" (or whatever heading).
• All three files share the same Pre-merge self-review loop phrasing (already aligned for primitive-mention as of scaffold: rename Windsurf surfaces to Antigravity (Google) #133).
• Pre-existing wording differences (Cursor "open a NEW composer tab" vs Copilot "open a NEW chat" vs Antigravity "Cascade thread") are intentional and stay tool-specific.