fix: export pyarrow tables and accept JSON exporter results

[peter-gy]

Exports failed when an output referenced a PyArrow table with string or dictionary columns. marimo's cell cache hashes Arrow data through a NumPy view, which rejects object-typed columns, and because that error happens in cache setup the failure surfaced as "failed in cell … with exception" with no type.

• When the native hash fails, owned graphs retry it with a digest of each referenced PyArrow value's type and IPC stream. Cells that hash natively keep their keys, and untracked graphs are unchanged.
• Exceptions from cache setup in owned graphs keep their type, so the error names it, e.g. "with TypeError".
• kind: native exports PyArrow Table and RecordBatch values as Arrow streams, like Polars frames. They previously fell back to marimo's pickle codec and were rejected.
• Custom exporters can return a JSON value as well as a BlobAsset. JSON results are stored as canonical marimo.json.v1, the same as kind: json, and read back through output.json().

The underlying hashing bug lives in marimo (is_data_primitive and _contiguous_tensor_bytes), so plain marimo sessions with the cell cache enabled still hit it.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[cubic-dev-ai]

All reported issues were addressed across 14 files

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>