Suggested bid and budget bounds

Resolve a human-readable query into the platform's opaque targeting ids used in the `TargetingSpec` (`countries`/`regions`/`cities`/`zips`/`metros` geo keys, and `interests`/`behaviors` entity ids) on `POST /v1/ads/create`, `POST /v1/ads/targeting/reach-estimate`, and `saved_targeting` audiences. The `dimension` param selects what is searched, `geo` (locations, further scoped by `geoType`), `interest`, `behavior`, `income`, `language` (Google-only), or the Meta-only work demographics `workPosition`, `workEmployer` and `workIndustry` (their ids feed `TargetingSpec.workPositions`/`workEmployers`/`workIndustries`). Availability of each dimension varies by platform (e.g. behaviours are Meta/TikTok only). Work industries are a fixed ~30-entry Meta catalog with no server-side query, so `workIndustry` matching, ranking and `limit` happen in Zernio. `language` is likewise a fixed, checked-in table of Google's targetable `language_constant` rows (id, ISO code, name) matched by name or code, capped at 20, with no network call; its ids feed `TargetingSpec.languages`. Results are normalized across platforms into a single shape, so the same client code consumes Meta, TikTok, LinkedIn, X, Pinterest, and Google results. TikTok geo searches return every matching level in one list (`type` is `country`, `region`, `city`, `district`, or `metro` for DMA areas) — `geoType` is not applied. Results are scoped to the advertiser's targetable markets, and every id is usable in `regions`/`cities`/`metros` keys on `POST /v1/ads/create`. LinkedIn geo searches also return every matching level in one list, and neither `geoType` nor `countryCode` is applied: LinkedIn's typeahead only returns a name and a URN per result, with no level or country field to filter on. Every result has `type` set to `location`, and its id is a `urn:li:geo:*` URN usable as a `regions[].key` on `POST /v1/ads/create`, `POST /v1/ads/boost` and `POST /v1/ads/targeting/reach-estimate`. Google geo searches resolve against Google's geoTargetConstants and return every matching level in one list; `geoType` is not applied (Google's `target_type` is an open taxonomy that does not map one-to-one onto the `geoType` enum), so filter client-side on the returned `type` (`country`, `region`, `city`, `zip`, `metro`, or the lowercased Google target type for rarer levels). `countryCode` scopes the search to one country. Each id is Google's numeric criterion id, usable as a `regions`/`cities`/`zips`/`metros` `key` on `POST /v1/ads/create`. Google city radius is not supported (pass a `customLocations` lat/lng pin for a radius); country targeting also accepts plain ISO codes via `countries` with no search call. Pinterest resolves against three whole-catalog endpoints (interests, locations, regions) with no server-side query or pagination, so matching, ranking and the `limit` cutoff all happen in Zernio; the catalog is independent of any ad account and results never carry `audienceSize`. Names come back localized to the connected Pinterest account's language (there is no way to force a locale), so match against whatever language that account returns. `geoType` routes to a different catalog: `country` and `metro_area` read the locations catalog (`type` is `country` or `metro`); `region` reads the regions catalog (`type` is `region`, its id a `regions[].key` on `POST /v1/ads/create`); `all` and the default `city` merge both catalogs with honest per-entry `type`s, since Pinterest has no city-level catalog and `city` is an alias for `all`, not a literal city search. `zip`, `subcity`, `neighborhood`, `place` and `geo_market` return a 400: Pinterest exposes no postal-code catalog, pass postal codes directly as `targeting.zips: [{ key }]` on `POST /v1/ads/create`. For geo queries, `q` should contain only the locality name (e.g. `"Amsterdam"`, not `"Amsterdam, NL"`). Use `countryCode` to disambiguate.