feat(agent): route background reviews via config

[kdmarble]

What changed

Adds optional background_review configuration for routing background memory/skill review agents to a separate
runtime.

By default, background reviews continue to inherit the active conversation model and provider. Users can now
configure a different provider, model, and max iteration count for those non-user-visible review turns.

Why

Background review agents can unintentionally compete with the active chat model/runtime. This allows users to
isolate that work onto a cheaper or separate endpoint without changing the visible chat session runtime.

How to test

scripts/run_tests.sh tests/run_agent/test_background_review_summary.py tests/run_agent/test_run_agent.py

Result:

312 passed

Also checked:

  python -m py_compile run_agent.py tests/run_agent/test_background_review_summary.py
  git diff --check upstream/main...HEAD

Platforms tested

• Linux
• Python 3.11.14

[alt-glitch]

Likely duplicate of #15508 — same configurable background review routing feature. Also overlaps with #13647.

[alt-glitch]

Likely duplicate of #15508

[kdmarble]

Closing in favor of consolidating around one of the other implementations. All three PRs solve the same problem (configurable background review routing) and there's no value in keeping competing approaches open. Happy to fold any unique contributions from this PR into whichever implementation maintainers prefer, or help review #13647 / #15508 directly.