Guardrails: Filter validator API #1256

Motivation

GET /api/v1/guardrails proxies the full validator catalogue from kaapi-guardrails (12 types today, including internal/experimental ones such as llm_critic, nsfw_text, llamaguard_7b, answer_relevance_custom_llm).

For the Glific guardrails v1.0 integration only four validators are productised and should be offered to tenants:

• topic_relevance_llm (Scope Control)
• uli_slur_match (Slur Match; replaces content_safety, v1.0 wants slur redaction only, no profanity pass)
• gender_assumption_bias (Gender Neutrality)
• pii_remover (Privacy Protection)

Exposing the rest in the list response lets UIs render validators we do not support yet.

Proposal

• Filter the upstream validators array in the list route to an explicit EXPOSED_VALIDATOR_TYPES set, preserving upstream order and leaving non-catalogue payloads (error bodies, empty responses) untouched.
• Render the OpenAPI example and the per-type reference in api/docs/guardrails_schemas.py from the same set so docs and response agree.
• Only the list is filtered; POST/PATCH /guardrails/validators/configs still forward any type upstream.
• Update docs/wiki/modules/guardrails.md.