On failure, btr_page_reorganize_low() fails to restore the old contents

In the function btr_page_reorganize_low(), there are several failure paths after the following:

  buf_block_t *old= buf_block_alloc();

Only right before the final return DB_SUCCESS we would release the allocated block back to the buffer pool:

  buf_block_free(old);

The implication is that if the page is found to be corrupted during a reorganize operation, we would leak a block that we allocated from the buffer pool. Eventually, this could lead to a hang of the server if we run out of the buffer pool.

alessandro.vetere recently fixed this in MDEV-40408, but only for MariaDB Server 11.8 onwards. The motivation of that fix was to improve performance, not to prevent a memory leak. A minimal fix will not change the allocation but just ensure that buf_block_free(old) will be invoked in every return path.

Furthermore, when a failure occurs, we fail to restore the contents of the page from old. We should do that in order to avoid making a bad situation worse. This follow-up to MDEV-18519 is needed to match the MySQL change Bug#37258813 btr_page_reorganize_low may cause crash recovery failure.