### The Feature

Add native support for **generic tag-based rate limiting**: a request-level tag identifies the caller (or any other logical entity), and the tag carries one or more configurable-window limits — **tokens, requests, or dollars** — enforced per routing attempt.

Concretely: - A chain/model-group declares one grouping per unit (`token_limits` / `request_limits` / `dollar_limits`), and **each individual limit entry carries its own `tag_id`** — so even limits of the same unit on the same chain can key by different tags (e.g. a `daily` token limit per `end_user_id`, a `weekly` token limit per `team_id`), not just different tags per unit. **The window is declared in raw seconds (`period_seconds`), not a fixed day/week/month enum** — this is deliberate: it makes genuine sub-day windows (e.g. an RPM-style "5 requests per 60 seconds" cap per identity) expressible with the exact same mechanism as a daily/weekly/monthly one, not a separate feature: ```yaml model_info: token_limits: limits: - name: daily tag_id: end_user_id limit: 500000 period_seconds: 86400 - name: weekly tag_id: team_id …