Batch load the merge requests for post manually merge check

This change improves how the system handles large numbers of merge requests by processing them in smaller batches of 500 instead of loading them all at once into memory. Previously, the code would fetch all open merge requests for a project simultaneously, which could cause performance issues or memory problems when dealing with projects that have many merge requests. Now it processes them in manageable chunks, making the operation more efficient and stable. Additionally, the target branch filter was changed from using a dynamic branch name to specifically target the 'master' branch.