Restore parallel makeMovie execution and add deterministic Dask token for containers

[haykh]

Motivation

• Ensure makeFrames continues to run via the executor (parallel) even when a data object is provided, as requested to avoid forcing a serial fallback while preserving laziness/xarray structures.
• Fix nondeterministic hashing/tokenization failures when container objects are captured by plotting workflows so parallel workers can reliably accept container handles.

Description

• Removed the in-process serial fallback so makeFrames submits all frames through the executor path and retained safe directory creation with os.makedirs(fpath, exist_ok=True) in nt2/plotters/export.py.
• Added BaseContainer.__dask_tokenize__ to nt2/containers/container.py to return a deterministic token tuple of (class name, path, reader.format.value) for stable hashing by Dask.
• Replaced/added tests: nt2/tests/test_export.py now asserts executor dispatch is used with data by injecting a fake executor, and nt2/tests/test_tokenize.py asserts deterministic container tokenization via dask.base.tokenize.

Testing

• Ran python -m compileall -q nt2, which succeeded.
• Ran pytest -q nt2/tests/test_export.py nt2/tests/test_tokenize.py, which failed before tests executed due to a ModuleNotFoundError for numpy during package import.
• Attempted python -m pip install numpy -q to unblock tests, but installation is blocked in this environment due to restricted network/package index access.

---

Codex Task