feat: multi-backend proxy with auto rate-limit jailing

Refactor from monolithic proxy to multi-backend architecture
(Backend interface: OpenCode Zen + Kilo gateway).

- Add auto rate-limit jailing: models returning HTTP 429 are
  hidden from /v1/models immediately and re-probed every 30 min
- Backend interface supports model aliases, custom headers, and
  per-backend routing
- Full Anthropic Messages API support with OpenAI format conversion
- 9 free models across two backends with Claude name aliases

internal/proxy/backend.go

@@ -0,0 +1,27 @@
+package proxy
+
+// Backend represents one upstream free-model provider (opencode zen, kilo, mimo, ...).
+//
+// Each backend knows its own model catalogue, how to resolve a requested model
+// name to the upstream ID, and the URL + headers required to forward a chat
+// completion request to it. The Proxy holds an ordered list of backends and
+// routes each request to the first one that recognises the model.
+type Backend interface {
+	// Name is a short stable identifier ("opencode", "kilo").
+	Name() string
+
+	// Models returns the free models this backend exposes, for /v1/models listing.
+	Models() []ModelInfo
+
+	// Resolve maps a client-requested model name to the upstream model ID.
+	// Returns ok=false when this backend does not recognise the model, so the
+	// router can try the next backend (or fall back to pass-through).
+	Resolve(model string) (resolved string, ok bool)
+
+	// ChatURL is the full upstream chat completions URL.
+	ChatURL() string
+
+	// Headers returns the headers required to authenticate/identify the
+	// request to the upstream (auth, client-id, request/session tracking, etc.).
+	Headers(requestID, sessionID string) map[string]string
+}