refactor(retry): server retry use tenacity

[salignatmoandal]

What

• Replaces the custom retry implementation in truss/templates/server/common/retry.py with tenacity.
• Adds tenacity to truss/templates/server/requirements.txt so the server runtime has the dependency when building/running the image.

This addresses the existing TODO in the file and aligns the server template with the rest of the codebase, which already uses tenacity (e.g. truss_handle, remote/baseten/service, control server).

How

• retry.py: Implemented the same retry behaviour using tenacity.Retrying with:
  • stop_after_attempt(count + 1) for the same total number of attempts (1 initial + count retries).
  • wait_fixed(gap_seconds) for the delay between attempts.
  • before_sleep callback to log the same messages as before ("Retrying..." on first failure, "Retry count: N" on subsequent ones).
  • reraise=True so the original exception is propagated after the last failure.
• requirements.txt: Added tenacity>=8.0.1 (aligned with the version in the project root pyproject.toml).

The public API of retry() is unchanged, so no updates were needed in model_wrapper.py.

Testing

• All existing tests pass without modification:
  • uv run pytest truss/tests/templates/server/common/test_retry.py -v — 5/5 passed.
• Behaviour is preserved: same number of attempts, same log messages, same exception propagation.