Modify Session temporary tablespace truncation on connection disconnect to reduce high CPU usage

Upstream:  https://bugs.mysql.com/bug.php?id=98869

With a large buffer pool, when thread-local intrinsic InnoDB tablespace is truncated at disconnection, the full LRU list scan leads to InnoDB choke. As a result, you will notice stall for transactions and High CPU usage spikes. 

Tested with PS 8.0.20

Reproduciable test case:

To repeat this, you need a large BP loaded with data init. In our test  following is the configuration,

innodb_flush_log_at_trx_commit=2
innodb_log_file_size=4G
innodb_buffer_pool_size=180G
innodb_flush_method=O_DIRECT

Loaded data using Sysbench  45 tables with 25M rows in each table.

To load data in innodb buffer run the following,
for i in $(seq 1 45); do ./use -e "SELECT * FROM sbtest.sbtest${i}"> /dev/null & if (( $i % 10 == 0 )); then wait; fi; done
As a result ,  innodb buffer size reached ~150GB

In the new session run,

1. select count(*) from information_schema.processlist; // Intrinsic Temporary Table
 or
 2. create a temporary innodb table explicitly. // External Temporary Table

disconnection and you will notice this issue.

The cpu usage is the result of `pidstat -u -p <mysql pid> 1`.
The stack is from `perf record -ag -p <pid>` and `perf report`.

Our perf result shows that mysql spent most of the time on at "buf_LRU_flush_or_remove_pages"  which same what we found in the user case.