feat: require explicit metrics and add scope flags

[adressop6789-max]

Pairs with asa-analytics!1584, which makes metrics required on POST /cli/metrics/.

Why the backend changed: subscribers and paid_subscribers are exact-uniqueness aggregations over profile ids, one hash set per entity per renewal window. On a keyword list that measured 11.4 GB against 0.13 GB for the same page without them, and production ClickHouse killed 17 queries on it. Omitting --metric used to mean "every column", which is how a paginated keyword export became the most expensive query on the cluster.

• --metric is now required, and its help says why a short list matters.
• New --app, --campaign, --ad-group scope filters go into the request body as app_id / campaign_id / ad_group_id. They are what unlocks subscribers on --entity keyword (the backend refuses those metrics account-wide with 422 cli_metric_scope_too_wide), and the cheapest way to make any keyword call fast.
• Agent docs and the flag table updated to match.

Ship to npm together with the backend: a client without --metric gets 422 after that deploy.

tsc -b clean, 174 mocha tests passing.