Protect HTML literal contents during translation retries

Make structural retries sentinel-free

Remap local sentinels after structural retries

Preserve sentinels during structural retries

Restore ordinary attention before smoke parity check

Align continuous batching smoke generation settings

Load TranslateGemma text model for continuous batching

Install exact Transformers runtime dependencies before translation

Align Hub dependency and add local runtime preflight

Import TranslateGemma auto classes from source-compatible module

Normalize GitHub slugs at clone boundary

Fix immutable model revision diagnostics and runner pin

Pass GitHub token securely to publishing Jobs

Pin runner and Job volumes

Add source-preserving Japanese documentation runner