Default expiration date is added to created table with a `TimePartitioning` when `expiration_ms` is explicitly none

Environment details

• OS type and version: Windows 10
• Python version: 3.9.0
• pip version: 21.1
• google-cloud-bigquery version: 2.13.1

Steps to reproduce / Code example

When running the following code a table is where the value of Partition Expiration is silently set to the default expiration date defined on the dataset.

from google.cloud.bigquery import (
    Table,
    SchemaField,
    TimePartitioning,
    TimePartitioningType,
)

table = Table(
    f"project.dataset.test_table",
    schema=[
        SchemaField("test", "STRING", "NULLABLE"),
        SchemaField("date", "TIMESTAMP"),
    ]
)
table.time_partitioning = TimePartitioning(
    TimePartitioningType.DAY, field="date", expiration_ms=None,
)

# bq being an already configured instance of google.cloud.bigquery.Client
bq.create_table(table)

To start it is good to note that the way to remove an expiration date (fully remove, not set to default) from a partition via the standard SQL is as follows:

ALTER TABLE ropo.orders 
SET OPTIONS (partition_expiration_days=NULL)

So even though you explicitly seem to set the expiration_ms to "no expiration date" in the code it will still silently remove the table after time x (depending on the dataset). The field expiration_ms only accepts None and positive integers, therefore, when the dataset does have one by default, there is no way to create the table with a TimePartition but without an expiration_date at all. Values like 0, null, and NULL will all throw an error.

I would advise changing the behavior as follows:

Set the default_values of some variable that is given the name DEFAULT_PARTITION_EXPIRATION_FROM_DATASET such that this behavior is clear from the function definition. Don't make the value of the variable None such that we still have this variable available to explicitly turn the expiration functionality off.