crimson/os/seastore: avoid new allocation when overwriting data in RBM for performance

crimson/os/seastore: avoid new allocation when overwriting data in RBM for performance

In 4K random write test, after seastore is filled up by 4MB extents,
current implementation performs deep copy in duplicate_for_write(), resulting in
significant performance degradation by 80%.
Therefore, this commit changes the deep copy behavior for bufferptr during the overwrite
situation to shallow copy, leaving the original data untouched.

Signed-off-by: Myoungwon Oh <myoungwon.oh@samsung.com>
Signed-off-by: Yingxin Cheng <yingxin.cheng@intel.com>
(cherry picked from commit 76b87855b4a6fb60a1fae59792aeebff3e8762d3)