There were two vulnerabilities in the previous fix(https://github.com/apache/doris/pull/20305):
1. `next_content` may not necessarily be a truly readable range
2. The last range of the merged data may be the hollow
This PR fundamentally solves the problem of reading amplification by rechecking the calculation range. According to the algorithm, there is only one possibility of generating read amplification, with only a small content of data within the 4k(`MIN_READ_SIZE `) range. However, 4k is generally the minimum IO size and there is no need for further segmentation.
d183e08f6d