chore(client): widen openai dep from ^1.30 to <3 (unblocks downstream openai 2.x migrations) (#812)

Unblocks downstream consumers that need openai 2.x.

The `scale-llm-engine` Python client does not use the `openai` package
at runtime. Verified via `grep -rnE '^(from openai |import openai)'
llmengine/` — zero matches. The only `openai.*` references are:

  - Local generated pydantic models in llmengine/data_types/gen/openai.py,
    imported as `.gen.openai` (NOT the external openai package)
  - URL strings in docstrings pointing at platform.openai.com
  - A comment noting api_engine.py was copied from openai-python (the
    code is self-contained, not a runtime import)

The `openai = "^1.30.0"` pin appears vestigial. Widening to
`>=1.30.0,<3` lets downstream packages upgrade to openai 2.x without
having to fork this client or wait on a coordinated bump.

Concrete downstream unblock:
  scaleapi/scaleapi egp-api-backend needs litellm 1.83.0 to clear
  CVE-2026-35030 (CRITICAL: OIDC /userinfo cache-key auth bypass),
  CVE-2026-35029 (HIGH), and GHSA-69x8-hrgq-fjj8 (HIGH). litellm 1.83
  requires openai>=2.8.0, which is currently blocked by this client's
  openai<2.0.0 cap.

Version bump:
  0.0.0.beta45 -> 0.0.0.beta46

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: scale-ballen