fix(docs-pipeline): say exactly how prompts were checked, state the docs-platform assumption

[darthrootbeer]

Context

The second cold review found the docs-pipeline said the maintainer read every prompt answer while its own run records say a Claude model graded them. It also never said up front that stages 2, 4b and 6 assume a ReadMe-synced repo, and the generic style guide imposed a house rule. Ticket: JBSR-099

Changes

• Every 'How these prompts were checked' paragraph (pipeline README, SETUP, ARCHITECTURE, general style guide, readability skill README) now says Sonnet 5.5 graded each answer against a good-answer list written before the run, and that the maintainer has not re-read every answer.
• README and SETUP state the ReadMe (readme.com) assumption for stages 2, 4b and 6.
• The 'end every doc with a prompt block' rule is marked optional in the style guide and in docs-style-check-structure, and mentioned under Knowledge sources.
• Lows: SETUP audience line, TICKET-1234 in README, smoke-test-run fence, readability README vault wording, starter KB date note in SETUP.
• No prompt wording changed, so no prompt was re-run. Offline checker and safety scan are clean.

🤖 Generated with Claude Code

https://claude.ai/code/session_01XV3Ggh86cf2xwUyz28XaN6

• Docs updated