fix: decode binary fields as UTF-8 text when possible

[gordonmurray]

Summary

• Binary/large_binary PyArrow columns containing valid UTF-8 are now displayed as readable text instead of base64
• Falls back to base64 encoding for actual binary data (non-UTF-8)
• Handles newer PyArrow versions (21.x) where .as_py() returns str directly instead of bytes

Test plan

• Verified string fields now show readable text (e.g. "New task") instead of base64
• Verified UUIDs, timestamps, and booleans render correctly
• Confirmed fallback to base64 is preserved for non-UTF-8 binary data