Context7 response cache and topic cache keys raise UnicodeEncodeError for a library id or query with a lone surrogate #756

Two more cache keys hash text with .encode(): Context7Client.get_library_docs (dokimi/dokimi/mcp/context7.py:137, f"{library_id}:{query}:{max_tokens}") and RagasSeedExtractor.extract_topics (dokimi/dokimi/questions/ragas_seed.py:100, f"{library_id}:{max_topics}:{max_tokens}"). The library id comes from --context7-id or, for a local: doc source, from the path; the query is a question text. Repro, with a cache directory set: Context7Client(cache_dir=d).get_library_docs('org/n\udcff', 'q') and get_library_docs('org/x', json.loads('"n\\ud800"')) raise UnicodeEncodeError: 'utf-8' codec can't encode character '\udcff' in position N: surrogates not allowed at context7.py:137; RagasSeedExtractor(mcp_client=..., cache_dir=d).extract_topics(library_id='local:n\udcff', library_name='n', max_topics=3) raises at ragas_seed.py:100. The second one also stops EvaluationPipeline._generate_questions end to end for a local: doc source whose directory name has a non-UTF-8 byte, after the pipeline's own topics-cache key (#716) is fixed. Fix: os.fsencode(...) for the library id (a path or argv value) and .encode('utf-8', 'surrogatepass') for the query (question text from JSON), which keeps every key that encodes today unchanged.

Found while fixing #716. The fix's PR is #761.