Truncate long string leaves when rendering body examples

Some media-type examples embed large blobs — e.g. GitHub's `repos.create_attestation` example is ~10KB of base64 (`canonicalizedBody`, `rawBytes`, `payload`), bloating `__doc__` for LLM consumption.

Follow-up: in `_examples_doc` (from #30), recursively cap string leaves at render time (e.g. >~100 chars → `prefix…+N chars`), keeping `OpSpec.body_examples` lossless. No content sniffing needed: capping leaf length handles base64 automatically while normal prose/URLs pass through. Optionally also cap examples shown per op.

Start at `_examples_doc` from #30 and trace how it renders `OpSpec.body_examples`. Recursively cap string leaves at roughly 100 characters while keeping `OpSpec.body_examples` lossless; verify that long embedded blobs are shortened and ordinary prose or URLs remain unchanged.

Written by the indexing model from the issue text.