They don't force. NTFS or 'database engine' don't care what happens with the storage system. It's up to the OS or storage system (HDD, RAID, SSD, NAS, etc) to deal with the write cache or reordering.
You're write about ordering, simply because IDE/ATA has never had queued barriers, TMK. But they care about caching.
1. NTFS == the OS. It has barriers to implement, to maintain the state of the data and metadata, which will include the occasional buffer flushing.
2. Write caching might not be able to be turned off, or bypassed. This has long been a problem with HDDs, since write caching allows much higher performance. So, the HDD will be lying when it returns that it flushed its buffers.
3. What DBMS doesn't care about what happens with the storage system? I would not want to ever have to rely on such a product, that's for sure.
Total apathy example. Even SQLite can be configured to ask for syncs all the time, to help ensure that data is where it expects it to be (Firefox does so, FI, and has had performance issues to deal with because of it, over the years). This is among the reasons why several SSD makers have advertised not using DRAM write caches, as that has long been an issue with HDDs (Crucial, Intel, and Samsung, right off): a successful write is one that actually happens, every time, rather than possibly being in a hidden queue to be written.