PS-7865 Dropping a table with discarded tablespace crashes the server

[voodam]

This is how the bug occurs:

1. If table have a lot of data, these conditions are satisfied and btr_search_update_block_hash_info returns true: https://github.com/percona/percona-server/blob/8.0/storage/innobase/btr/btr0sea.cc#L588 on one of LOAD DATA commands execution. This causes the creation of a hash index.
2. On discarding table this line is reached: https://github.com/percona/percona-server/blob/8.0/storage/innobase/row/row0mysql.cc#L3994. This will be the problematic index on which assert fails because of index->space == FIL_NULL but block->page.id.space() stays the same.
3. This erros occurs only on a lot amount of data, otherwise a hash index will not be created and hence will not be droped.

I suppose that solution can be in widening assert condition to include index->space == FIL_NULL case. But I'am not sure, is it okay that we drop the reseted index in btr_search_drop_page_hash_index? But early returning from the function probably is not an option, because in that case assert_block_ahi_empty(block) is failed:

  if (block->index == nullptr ||
      block->index->space == FIL_NULL) {
    rw_lock_s_unlock(latch);
    assert_block_ahi_empty(block);
    return;
  }

Assert was originally added in percona@195760a57028 commit. Probably it's not good idea to widen other same asserts because other cases are not related with dropping index, but with creating, updating etc. But I'am also not sure.

I added a little faster version of a testcase. It differs from the original one, but works in the same way.