`try_cast_literal_to_type` doesn't respect timestamp precision

[vrongmeal]

Describe the bug

In the function try_cast_literal_to_type: https://github.com/apache/arrow-datafusion/blob/928662bb12d915aef83abba1312392d25770f68f/datafusion/optimizer/src/unwrap_cast_in_comparison.rs#L295

Converting from Timestamp(Nanosecond, _) to Timestamp(Microsecond, _) doesn't actually return the correct value. It simply interchanges the inner integer.

To Reproduce

Running the following:

#[test]
fn test_repro() {
    let new_scalar = try_cast_literal_to_type(
        &ScalarValue::TimestampNanosecond(Some(123456), None),
        &DataType::Timestamp(TimeUnit::Microsecond, None)
    ).unwrap().unwrap();

    // This passes which is wrong
    assert_eq!(new_scalar, ScalarValue::TimestampMicrosecond(Some(123456), None));

    // This fails!
    // assert_eq!(new_scalar, ScalarValue::TimestampMicrosecond(Some(123), None));
}

Expected behavior

// This should be the correct result
assert_eq!(new_scalar, ScalarValue::TimestampMicrosecond(Some(123), None));

Additional context

[comphead]

❯ select extract(microsecond from 123456::timestamp);
+---------------------------------------------+
| datepart(Utf8("MICROSECOND"),Int64(123456)) |
+---------------------------------------------+
| 123.456                                     |
+---------------------------------------------+

The sql repr works better.

Thanks @vrongmeal for reporting the bug.