Today the file `-o create.json` writes only the final AI-progress
payload — no `collector_id`, no name, no view_url. The documented
recipe in references/recipes.md depends on jq reading the
collector_id out of that file:

    COLLECTOR_ID=$(jq -r '.collector_id // .id' create.json)
    bdata scraper run "$COLLECTOR_ID" ...

Today that returns the string "null" because the field doesn't exist
in the file. Every script that follows the docs to chain create →
run is silently broken.

This change wraps every termination path (success, AI-trigger
failure, status=failed, polling exception) in one envelope:

    {
      "collector_id":    "c_...",
      "name":            "audit-r4-...",
      "status":          "done" | "failed" | "ai_trigger_failed" | "poll_failed",
      "completed_steps": [...],
      "view_url":        "https://brightdata.com/cp/scrapers/c_...",
      "created_at":      "2026-05-18T07:28:30Z",
      "error":           "..."   // failure paths only
    }

Notable design choices:

* Every termination path writes the same shape, including failure
  paths that previously wrote nothing. So a script using
  `jq -r '.collector_id'` always recovers an id when one exists —
  even from a stub collector that hit the AI-Flow parallel-job cap.
  This makes good on SKILL.md's promise that every failure path
  surfaces the collector_id.
* `view_url` is included on every envelope so the user has a one-
  click recovery path to inspect / finish / delete the scraper in
  the dashboard, without needing to know the URL pattern.
* `created_at` is taken from the template-creation response when
  the API provides it (`Create_template_response.created`),
  omitted otherwise — never invented.
* New `--legacy-output` flag preserves today's bare-progress shape
  for one minor version so any existing scripts that depended on
  the old shape have a migration window. Slated for removal in
  the next major.
* Stdout (the success summary printed to TTY) is unchanged. Only
  the machine-readable `-o` / `--json` / `--pretty` payload is
  reshaped.
* Scoped strictly to `src/commands/scraper.ts` and the new
  envelope type. The shared HTTP client and other commands
  (scrape, search, discover, pipelines, browser) are untouched.

Tests: 4 new `build_create_envelope` unit cases covering success,
omitted-created_at, failure-with-error, and view_url-on-every-
path. 5 new `handle_create_scraper` integration cases covering
success envelope, the documented jq recipe, --legacy-output
preserving the bare shape, AI-trigger failure envelope (the
stub-collector recovery path), poll-status-failed envelope, and
poll-exception envelope. Two existing tests updated from strict
opts-object matches to objectContaining-style (the contract is
now the envelope shape, not the bare payload).

55 / 55 scraper tests pass. The 9 pre-existing failures in
unrelated suites (daemon, add-mcp, browser, discover, scrape) on
main are unchanged by this PR.

Spec: brightdata/skills repo, proposal at
skills/scraper-studio/proposals/PR-2-create-envelope.md (to be
filed alongside this PR).