Feature request: Allow read_items() to accept partition/sort key names instead of always calling DescribeTable

[gautamp-aiml]

Is your feature request related to a problem? Please describe.
Yes. In high-throughput DynamoDB workloads (e.g., AWS Lambda processing Kinesis streams), every call to wr.dynamodb.read_items() triggers a DescribeTable API call to fetch the key schema.
This generates a huge number of CloudTrail events, which in turn drives up CloudTrail and GuardDuty costs.
When the table schema is already known (partition key + optional sort key), these extra API calls are unnecessary and expensive.

---

Describe the solution you'd like
I would like read_items() to support explicitly passing partition and sort key names.
If these parameters are provided, Wrangler should skip the DescribeTable call and use the provided schema directly.
The default behavior (auto-discover schema via DescribeTable) can remain for backwards compatibility.

Example API shape:

df = wr.dynamodb.read_items(
    table_name="my-table",
    partition_key="pk",
    sort_key="sk",
    partition_values=["pv1", "pv2"],
    sort_values=["sv1", "sv2"]
)

---

Describe alternatives you've considered

• Monkey-patching read_items to skip DescribeTable. This works but is fragile and version-specific.
• Replacing Wrangler with raw boto3 calls and custom DataFrame conversion. This avoids the issue but loses Wrangler’s ergonomics.
• Adding caching around DescribeTable to reduce the frequency, but this still adds complexity and doesn’t eliminate unnecessary API calls.

---

Additional context
This feature would make read_items safer and more cost-efficient in serverless, high-volume architectures, where cold starts and container churn can otherwise generate thousands of unnecessary DescribeTable calls.
It would also give developers more control over performance and cost tradeoffs, while keeping backward compatibility for existing users.

---