test: finish HTTP mock processor coverage

[kitlangton]

Summary

• finish the processor effect migration so the remaining processor coverage runs through the real LLM service and HTTP mock server
• extend the mock LLM server with a request-level reply DSL for SSE flows, including reasoning chunks, pending tool hangs, usage chunks, and connection resets for retry coverage
• remove the remaining in-process reasoning shim from processor-effect.test.ts and keep the common test call sites on the short llm.text/error/toolHang/hang helpers