docs(scraper-studio): document 3-job cap, propose PR-11 backoff + stu…

[anil-bd]

…b asks

The AI Flow caps concurrent scraper create generations per account (currently 3). Today the cap is undocumented in this skill, in the bdata scraper --help output, and in api-flow.md — users discover it by getting 429s in production.

This commit lands the documentation half of PR-11:

• SKILL.md "Common mistakes" gets entry 9 naming the cap and pointing at the new CLI flags (--max-retries, --no-retry) that handle it.
• SKILL.md "Troubleshooting" gets two new rows: one for the expected in-progress backoff stderr line, one for the retries-exhausted case (actionable, points at the dashboard).
• references/api-flow.md documents the cap inline on the /dca/collectors/{id}/automate_template endpoint and notes the stub-collector pathology (template POST always succeeds, so a 429 here leaves a half-built collector behind).
• New proposal at proposals/PR-11-backoff.md captures the full story: Half A (client-side auto-backoff + stub-recovery note — shipped in the cli repo) and Half B (open server-side asks — reject template POST when cap is full, expose DELETE /dca/collector/{id}).

No content change to prompts.md, recipes.md, or the rest of SKILL.md.

The CLI side ships separately on cli@feat/scraper-create-429-backoff.