review InnoDB page fill factor

Tests undertaken in PERF-206 show significant performance improvements of 30% for sysbench-tpcc when INSERT uses a column default value instead of NULL. The reason is that the row is UPDATEd later, filling proper values for the columns inserted as NULL. This leads to a growing row image which eventually leads to a page split. The page split in turn leads to mutex contention.

The issue has been observed for MySQL as well, as explained here

We should review if the hardcoded page fill factor of 15/16 for InnoDB can be chosen better. Other DBMS use a lower fill factor of i.e.70% (PostgreSQL) or make it configurable (Oracle DB, configurable per table).