MwCAS implementation tracking

[XiangpengHao]

This issue tracks the modifications to current MwCAS implementation.

The continuous benchmark can be found here: https://dev.azure.com/haoxiangpeng/pmwcas/_dashboards/dashboard/7ed85ecc-8da6-4958-977e-efb77c25afcc

• Address flushes are added back because they are believed to be necessary, this result in performance drop about:
  

• https://github.com/XiangpengHao/pmwcas/blob/dev/src/mwcas/mwcas.cc#L708 is deleted, resulting in performance increase by:
  

• After a second thought, the address flush after the description installation should not be necessary. So remove it and we get back to the old performance:
  

• Now we use RTM by default and use the normal CAS based descriptor installation as the fallback plan.

• We now make a copy of the data before persisting it, by doing this we can avoid some cache misses.