[bug] AvailableNow trigger causes stream to stop after one batch

The availableNow trigger appears to complete after a single batch of SQS messages is consumed, even if there are plenty of messages on the queue:

```
df = (spark.readStream.format("s3-connector")
.option("spark.s3conn.queueRegion": "us-east-1")
.option("spark.s3conn.fileFormat": "parquet")
.option("spark.s3conn.maxFilesPerTrigger": "2048")
.option("spark.s3conn.queueUrl": "https://sqs.us-east-1.amazonaws.com//")
.option("spark.s3conn.queueFetchWaitTimeoutSeconds": "10")
.option("spark.s3conn.sqs.longPollingWaitTimeSeconds": "5")
.option("spark.s3conn.sqs.visibilityTimeoutSeconds": "600")
.option("spark.s3conn.pathGlobFilter": "*.parquet.gz")
.schema(schema)
.readStream("s3:...."))

df.writeStream.trigger(availableNow=True).foreachBatch(stream_callback)
.option("checkpointLocation", )
```

This was reproduced on Databricks DBR 14.3

Reproduce the supplied SQS setup on Databricks DBR 14.3 and inspect the connector path exercised by readStream with trigger(availableNow=True). There is no source file or test named in the issue; done means queued SQS batches continue to be consumed until the queue is exhausted rather than stopping after one batch.

Written by the indexing model from the issue text.