Fix round-trip write/read issues from Parquet/CSV/JSON

[jaychia]

This long-running issue records any bugs found with a roundtrip write + read from formats such as Parquet and CSV.

Tests were added here: #1616

Parquet

• Reading FixedSizeList type is broken (looks like a bug in arrow2)
• Reading timezone-aware columns is broken See: [BUG] Fix parquet timestamp tz roundtrip inference #1625
• Reading multiple files with the Tensor type is broken

CSV

• Reading the Decimal type from a CSV with floats in the column is broken See: [BUG] Fix CSV roundtrip for decimals (actually an f64->decimal casting bug) #1626

[murex971]

@jaychia the test here marked as skipped

Daft/tests/io/test_parquet_roundtrip.py

Line 107 in d2f28d6

@pytest.mark.skip(reason="Currently fails when reading multiple Parquet files with tensor types")

is working fine, maybe it has been resolved.

[jaychia]

@jaychia the test here marked as skipped

Daft/tests/io/test_parquet_roundtrip.py

Line 107 in d2f28d6

@pytest.mark.skip(reason="Currently fails when reading multiple Parquet files with tensor types")

is working fine, maybe it has been resolved.

Oh yes -- good point! Interesting, something seems to have fixed it in the meantime :)

Removed the skipping:

#2024