v0.12.1: adopt gemini-3.5-flash GA, drop obsolete cache fallback

v0.12.0: 3.1 image/TTS models, google-genai 2.x

feat: improve retry backoff and update Gemini models

Align retry config across server.py and index.py: switch to
wait_exponential_jitter (initial=2s, max=60s, jitter=5s) to prevent
thundering herd on 429/503. Increase index.py retries from 3 to 5 and
expand to catch ServiceUnavailable + InternalServerError (not just
ResourceExhausted). Add before_sleep_log to index.py for retry visibility.
Fix _rebuild_sync to catch per-file errors instead of crashing.

Update models: Pro to gemini-3.1-pro-preview, Nano Banana Pro to
gemini-3-pro-image-preview, add speech-fast alias for
gemini-2.5-flash-preview-tts. Add model param to generate_speech tool.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

docs: uncomment config.toml examples for clarity

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

v0.4.0: integration tests, model auto-discovery, gemini-flash-latest