InnoDB may corrupt its log before upgrading it on startup

In the function srv_start(), there are some optional adjustments for MDEV-11623 as well as something related to innodb_data_file_path. These adjustments would write log records, even though the log has not yet been upgraded to the configured format (create_log_file_rename()).

For example, killing the server at an unfortunate time during an upgrade from 10.1 could theoretically make the data directory unrecoverable.

I think that we must try to cover this code in our regression tests. There should already be a debug assertion for catching writes to a wrong-format log file.