server: implement GET /v1/models for gateway model discovery

[kevinsslin]

Fixes #60.

Problem

The README quickstart recommends CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY=1, but there is no /v1/models route, so Claude Code's startup discovery request (GET /v1/models?limit=1000) falls through and the flag silently does nothing.

Change

Single concern: add GET /v1/models, serving the registry's model list in the Anthropic models-list shape.

• data[].id + data[].display_name ("<model> (<provider>)") + type: "model"
• has_more, first_id, last_id; limit honored, unknown query params tolerated
• The Anthropic-style aliases from the registry are included. Per the official gateway protocol (https://code.claude.com/docs/en/llm-gateway-protocol), Claude Code ignores discovered ids that don't start with claude/anthropic, so the aliases are what actually surface in the /model picker; raw provider ids remain listed for other Anthropic-compatible clients.
• No auth on the endpoint, matching the server's existing no-inbound-auth posture; response is static registry data (fits Claude Code's 3s discovery timeout, no redirects).

Testing

• 4 new integration tests in tests/server.rs (list shape, claude-prefixed aliases present, limit + has_more, unknown query params tolerated)
• cargo test: 599 passed, 0 failed
• cargo fmt --all --check clean
• cargo clippy --all-targets -- -D warnings: identical pre-existing error count before/after (34), none in touched files
• Live check on a local build:

$ curl -s 'http://127.0.0.1:18799/v1/models?limit=1000' | jq '[.data[].id | select(startswith("claude-"))]'
["claude-fable-5","claude-haiku-4-5","claude-haiku-4-5-20251001","claude-opus-4-7","claude-opus-4-8","claude-sonnet-4-6","claude-sonnet-5"]

[raine]

Thank you