Add model benchmarking and cleanup examples

[KhanN-byte]

What changed

• Added examples/benchmark.py to compare installed models using median time to first token, total duration, load duration, prompt throughput, and generation throughput.
• Added optional warm-up runs, reproducible generation settings, selected-model execution, and CSV summary output.
• Added examples/modelDetection.py to list installed models and their reported storage usage, select models interactively, and remove them after explicit confirmation.
• Added cancellation defaults, input validation, per-model error handling, and a warning that shared model layers can affect actual freed space.

Why

These examples make it easier to understand local model performance on a particular machine and safely manage disk space used by installed models.

User impact

The changes add standalone examples only and do not alter the Ollama Python client API. Model removal is never performed without the user typing yes at the confirmation prompt.

Validation

• pytest -q ollama tests: 98 passed
• ruff check examples/benchmark.py examples/modelDetection.py: passed
• ruff format --check examples/benchmark.py examples/modelDetection.py: passed
• python3 -m py_compile examples/benchmark.py examples/modelDetection.py: passed
• Benchmark smoke-tested against qwen3.5:4b
• Model manager smoke-tested by listing installed models and cancelling without deletion