Bug #104639 Simplify the purge_sys->read_view creation to reduce mutex contention

Description:
Currently the purge_sys->read_view is copied from oldest read view or create a new one by taking the trx_sys->mutex lock, so heavy work loading will block purge threads to reclaim the undo space.

How to repeat:
run benchmark, and check the purge progress.

Suggested fix:
We don't need to copy the whole read view or create a new one by copy the  trx_sys->rw_trx_ids array.

Actually we just need to get min(min_active_id - 1, oldest_view->low_limit_no) as the upper_limit_id/low_limit_no/low_limit_id of purge_sys->read_view, by maintain a atomic variables of the oldest read view's low_limit_no, we can skip acquiring of the trx_sys->mutex.

Which will make purge faster than before.

Hi Mr. Flou,

Thank you for your performance improvement request.

We consider that it has sufficient merit to be considered in more detail.

Verified as the performance improvement request.

Just setting the version correctly.