feat(ai): add packages/ai provider layer (Chunk #8)

Self-contained SSE-based provider layer for OpenAI-compatible endpoints
(Ollama/llama.cpp/LM Studio/vLLM), Anthropic, Gemini, and OpenRouter.

- budget.ts: re-export core token-budget engine (countTokens,
  allocateBudget, DEFAULT_BUDGET)
- client.ts: streamCompletion/collectStream/complete with budget
  enforcement via enforceBudget + measureBudget
- discover.ts: auto-detect local providers by probing common /v1/models
- adapters.ts: per-provider param mapping + streaming adapters + factory
- thinking.ts: thinking/reasoning-param normalization per provider
- http.ts: SSE parsing + 429 retry
- errors.ts: BudgetExceededError

35 unit/integration tests pass; typecheck and lint clean.

packages/ai/src/budget.ts

@@ -0,0 +1,15 @@
+/**
+ * Token budget re-exports.
+ *
+ * The budget engine lives in `packages/core` (the single source of truth for
+ * token accounting). This module re-exports it so the AI layer has a single
+ * import surface, per CHUNK#8.md §3.
+ */
+
+export {
+  countTokens,
+  allocateBudget,
+  DEFAULT_BUDGET,
+} from "@pubscript/core";
+
+export type { Budget } from "@pubscript/core";