WL#11348: Defaults: Increase Slave's Multi-Threaded Event Applier Buffer

Affects: Server-8.0   —   Status: Complete

This worklog changes the default of slave_pending_jobs_size_max from 16M to 128M.

The default value for the variable being discussed here is:
It's not uncommon for users to have some large transactions mixed in with more
"normal" smaller ones.
If these large transactions are => slave_pending_jobs_size_max in size then they
completely clog up the MTS applier pipeline and the slave will begin lagging.
For example, see: https://forums.mysql.com/read.php?26,660062,660074#msg-660074
The current default of 16M is unnecessarily small given today's memory sizes and
our other defaults.
Given all of this, we should increase the default size from 16M to 128M.
FR1: The output of '@@global.slave_pending_jobs_size_max' is 128M by default.

NFR1: These default changes must not cause performance regression
      in SysQA perf tests.
1. the default of slave_pending_jobs_size_max is changed from 16M to 128M
   for avoiding some big transactions clog up the MTS applier pipeline
   and keeping high concurrency of slave workers, and it should be bigger
   than the default of max_allowed_packet.
2. Cross-version Replication Impact
     - No risk for cross-version replication.
3. Upgrade and Default values
     - No risk for upgrades on default configs (MTS is disabled).
     - Risk for upgrades for people with MTS enabled on default
       configs (memory usage may be a bit higher)
     - No risk for upgrades with people with MTS enabled and
       slave_max_pending_jobs_size set
4. Cross-platform Impact
     - No cross-platform impact
5. Resource Usage Side-effects
     - The max memory used by all applier worker queues combined together
       may grow up to 128 MB (Previous default was 12MB). NOTE: docs can
       adjust the manual if needed and eventually the user can calculate
       the resources it needs to deploy the MySQL server (taking into
       account slave_max_pending_jobs_size buffer size).
6. For the QA portion, we should have the standard SysQA tests run as
   has been done for all other 5.7 and 8.0 defaults change worklogs,
   to check for any unexpected performance regressions or other added