Stale and legacy material sweep in bindings/python and the root README #417

A sweep of bindings/python and the repository docs for stale and legacy material found little, because the docs are kept close to the code: the strict mkdocs build passes, every example has a docs page and a nav entry and the reverse, every documented example flag exists, every file path and environment variable named in the docs exists, and every mentioned workflow file exists.

What is stale, and removed by the linked PR: the root README.md states a wheel of about 62MB download and 87MB installed, while the installation page measures 70.2 MiB and 97.8 MiB for 26.10.2. scripts/profile-python/README.md and results.md tell the reader to run the profiler with bare python, which breaks the uv-only rule; the profiler itself still runs. Three symbols have no caller anywhere in the repository: _java_class_name in results.py, _UNSET in type_conversion.py and DEFAULT_BATCH_SIZE in the internal dataframe.py. test_server_context_manager exists twice (test_server.py and test_server_patterns.py); the test_server.py copy only asserts is_started() after a time.sleep(1) and the other one checks the full lifecycle.

Found and left alone: files owned by open PRs (#392 ensure-build-tools.sh is called by nothing, testing.md lacks rows for four test files, test_jvm.py has an unused _jvm_running helper, uv pip and pip install in example docker wrappers), the Java ColumnBatcher.parseColumnSpec legacy semicolon form (Python always sends JSON; dropping it needs a jar rebuild and the benchmark build scripts pair jars and Python sides across versions), and the benchmark harness.