feat(web): add Serply as an optional backend for web and Scholar search

[googio]

What

Adds Serply as an optional keyed backend in the web layer, following the pattern Tavily already uses (key from config or env, nothing changes without it):

• WebSearchClient tries Serply's /v1/search/ (Google web results) after Tavily and before the DuckDuckGo HTML fallback.
• GoogleScholarClient queries Serply's /v1/scholar/ first when a key is set and falls back to scholarly if that call fails. The client can now be constructed without scholarly installed as long as a Serply key is present.
• WebSearchConfig gains serply_api_key / serply_api_key_env (default SERPLY_API_KEY); _literature.py passes the resolved key to WebSearchAgent the same way it does for Tavily.

Plain urllib only, no new dependency.

Why

Refs #321. There is currently no keyed option for Google web or Google Scholar results, and Scholar results require the scholarly scraper, which Google blocks on unattended runs. Serply gives Stage 4's web augmentation a stable source for both, and lets Scholar work without the optional web extra.

Notes

• Serply support is optional, and behavior is unchanged when SERPLY_API_KEY is unset.
• Requests carry an explicit User-Agent; Serply's edge rejects requests without one.
• Venue and year for Scholar results are parsed from Serply's description line ("A Author, B Author - Venue, 2021"); a trailing , YYYY is stripped so venues such as 2022 IEEE/CVF Conference ... (CVPR), 2022 keep their name.
• Tests added in tests/test_web_search.py (TestSerplyBackend), tests/test_web_scholar.py (TestSerplyScholarBackend) and tests/test_rc_config.py; tests/test_rc_executor.py extended for the new kwarg. Network is mocked.
• README web_search example gains the serply_api_key_env line. Translated READMEs under docs/ were left alone.

Testing

$ env -u SERPLY_API_KEY -u TAVILY_API_KEY .venv/bin/python -m pytest tests/test_web_search.py tests/test_web_scholar.py tests/test_rc_config.py -q
................................................................         [100%]
64 passed in 0.27s

$ env -u SERPLY_API_KEY -u TAVILY_API_KEY .venv/bin/python -m pytest tests/ -q --ignore=tests/e2e_docker_sandbox.py --ignore=tests/e2e_real_llm.py
2965 passed, 56 skipped, 1 warning in 82.23s (0:01:22)

Live check with a real SERPLY_API_KEY and scholarly not installed: WebSearchClient(serply_api_key=...).search("knowledge distillation survey 2024") returned 5 results (source serply, 1.30s); GoogleScholarClient(serply_api_key=...).search(..., limit=5) returned 5 papers with authors, year, venue and citation counts (first: gou2021knowledge, OpenAlex id W3034368386); WebSearchAgent(serply_api_key=...) produced 3 web and 3 scholar results. With no keys set, the client goes to the DuckDuckGo path exactly as before.

Disclosure: I work with Serply. Happy to adjust scope or naming, or drop this
entirely if it isn't a direction you want for the project.