[Ticket]: Validate the Anthropic-endpoint capabilities of the Envoy AI Gateway

### Type

Spike / Investigation

### Summary

We need to validate the Envoy AI Gateway's Anthropic-endpoint (`/v1/messages`) capabilities — and catalogue which models can be called through it — because we want to fully support clients using Claude Code, which speaks the Anthropic API.

Expected result:
> A catalogue of every model callable via the Anthropic endpoint, plus a decision: if too few, implement a model proxy so Claude-Code clients have full coverage.

### Intent

Make Claude Code (and other Anthropic-API clients) first-class against our gateway.

### Source of truth (links)

- AI Governance sprint (`Sprint.txt`).
- Gateway: `charts/core-gateway`, `charts/ai-models` (Envoy AI Gateway, AIEG). Runtime/gateway epic: #143.
- Anthropic API: `/v1/messages` shape.

### Current Behavior

It is unclear which models work through the Anthropic endpoint; Claude-Code support is unverified.

### Expected Behavior

We know exactly which models are Anthropic-callable; gaps are closed (proxy) if needed.

### Acceptance Criteria

- [ ] Every model callable via the Anthropic endpoint is catalogued (with evidence of a working call).
- [ ] Claude Code is tested end-to-end against the gateway.
- [ ] If the callable set is too small, a model proxy is designed/prototyped to widen coverage.
- [ ] Verification evidence (working Anthropic-endpoint calls + a Claude Code session) is provided.

### Out of Scope

- Full productionisation of a proxy (prototype + recommendation if needed).

### Technical Context

AIEG translates OpenAI/Anthropic schemas per model/provider; some models may only expose OpenAI-compatible routes → a translation proxy may be needed.

### Risks

- Few models support Anthropic natively → proxy becomes load-bearing for Claude Code.

### Test Plan

- Call each model via `/v1/messages`; run Claude Code against the gateway; record results.

### Verification evidence

_To be filled on completion: model catalogue + Claude Code session evidence._

### Human accountable owner

@Koufan-De-King

### AI Usage Declaration

- [x] Drafting the ticket
- [ ] Not used
- [x] I have declared AI usage above (ticked the relevant items, or "Not used").

### Human verification completed

- [ ] I understood the intent
- [ ] I checked the source of truth
- [ ] I reviewed all AI-generated text/code
- [ ] I verified the implementation manually
- [x] I checked for hallucinated assumptions
- [x] I am the accountable owner and accept responsibility for this ticket.

Review Sprint.txt and the charts/core-gateway and charts/ai-models directories first, then inspect the gateway's Anthropic /v1/messages path. Call each configured model through that endpoint and run Claude Code end-to-end, recording working evidence. Done means a complete model catalogue, session evidence, and a proxy prototype and recommendation if coverage is insufficient.

Written by the indexing model from the issue text.