[leo_storage] Avoid write operations at more earlier timing than #592 do

[yoshiyuki.kanno]

This issue is a follow up for #592.

As reported on #1156 (comment), #592 worked as we intended and succeeded in avoiding that the disk full happens however it turned out that the compaction on LeoStorage which available disk space is really small took so much time as the less available disk space there is, the more time any disk operations (seek/read/write) tend to take (because there are barely physically continuous disk blocks).

That being said, we may have to avoid write operations more earlier timing at which the compaction throughput isn't degraded.

NEED BENCHMARK.