fix(litellm): edge rate limit + HSTS on the public LiteLLM route

[Mushtaq-BGA]

What

In litellm mode the litellm.<domain> route publishes the LiteLLM UI and its admin API to the internet — and the UI must stay reachable so users can log in. Path-restricting the route can't reduce that surface (the UI is a client of the admin API), so this PR hardens the route with two edge controls instead:

• HSTS — Strict-Transport-Security response header (via a Gateway API ResponseHeaderModifier filter on the route) so a browser/SDK never sends the UI login password or a virtual key over plaintext :80 before the 301 to HTTPS.
• Local per-client-IP rate limit — an Envoy Gateway BackendTrafficPolicy (type: Local, no Redis / external ratelimit service) with a Distinct match on x-forwarded-for, giving each source IP its own counter. Blunts automated brute-force against the login/admin API and caps runaway abuse.

Notes

• Both are toggleable and tunable via roles/litellm/defaults/main.yaml (litellm_hsts_*, litellm_rate_limit_*).
• Idempotent / self-healing: HSTS is built as a fact so disabling it re-renders the route with no filter; the rate-limit policy has a paired state: absent task when disabled, and is removed on teardown.
• Coarse by design. Local rate limit is per-gateway-replica, and clients behind a shared NAT/egress IP share one counter. Per-identity limits + proper multi-user login (OIDC/SSO) come with the in-progress auth-capability work.
• Also corrects stale /key/verify extAuth references in the role docs — superseded by the litellm-mode redesign (gateway does TLS+routing only; LiteLLM is the sole auth boundary).

Validation

• YAML parses; HSTS filter fact renders correctly for enabled/disabled.
• Rendered HTTPRoute (with filter) and BackendTrafficPolicy both pass kubectl apply --dry-run=server against the live Envoy Gateway v1.7.1 CRDs.

Companion to the litellm-mode auth-scoping PRs (this repo #12; landing #17).