fix: retry OpenAI streaming rate limits

[Ben Younes (ousamabenyounes)]

Summary

• add a retrying fetch wrapper for ChatOpenAI-backed providers so retryable HTTP responses from the OpenAI SDK transport are retried even on streaming runs
• keep one retry budget for OpenAI-backed transports by disabling the outer LangChain retry layer where the fetch wrapper owns retryable HTTP and transport failures
• preserve existing custom fetch transports for ChatGPT OAuth and Vertex MaaS by wrapping them instead of replacing them
• document that OpenAI/OpenAI-compatible streaming HTTP responses are covered by provider retry attempts

Fixes #431

Test verification (RED → GREEN)

RED on origin/main with the new regression test only:

FAIL  test/openai-provider-retry-fetch.test.ts > createModel OpenAI provider retry fetch > retries provider rate-limit responses through the OpenAI SDK fetch
AssertionError: expected undefined to be type of 'function'

Expected: "function"
Received: "undefined"

GREEN on this branch:

pnpm vitest run test/openai-provider-retry-fetch.test.ts
Test Files  1 passed (1)
Tests  3 passed (3)

Additional validation on this branch:

./run-ci.sh
install: pnpm install --frozen-lockfile
format: pnpm run format:check
lint: pnpm run lint:check
typecheck: pnpm run typecheck
build: pnpm run build
cli-smoke: OPENWIKI_DEV=1 node dist/cli.js code --dry-run --init
test: pnpm test
Test Files  54 passed (54)
Tests  662 passed (662)