Problem
-------
The transaction dependency writeset history map doesn't
have protection from concurrent access.

Analysis
--------
This is because we are not locking m_writeset_history by
LOCK_slave_trans_dep_tracker at all places wherever it is
accessed whereas the same object is getting locked while
clearing the history map.

Fix
---
The Writeset_history::m_writeset_history is now guarded by
LOCK_slave_trans_dep_tracker where ever it is accessed and
this protects concurrent update on Writeset_history::
m_writeset_history.

Note : This bug fix also protects m_opt_tracking_mode variable
       explicitly using atomics.

RB#24250