ADBDEV-4648: Limit diskquota hash table's size according initial request

[KnightMurloc]

Diskquota did not control the size of its hash tables in shared memory and could
have consumed shared memory not intended for it, potentially impacting the other
database subsystems. Hash tables can also grow indefinitely if the background
process on the coordinator has not started, which can happen for a number of
reasons: gone done with error, pause, isn’t started. In this case, data is not
collected from segments and some hash tables (active_tables_map, relation_cache,
relid_cache) would not be cleared and would overflow.

This patch adds a limit on the size of all hash tables in shared memory by
adding a function that checks whether the hash table is full. The function
returns HASH_FIND if the map is full and HASH_ENTER otherwise. It also report a
warning if the table is full. Implemented a GUC that controls how frequently the
warning will be reported, as it could be reported too frequently. Also added a
GUC to control size of local reject map. The size of global reject map is set
as diskquota_max_local_reject_entries * diskquota_max_monitored_databases.

The test_active_table_limit test has been changed. Firstly, the value of
max_active_tables was changed from 2 to 5, since tables from all databases
processed by diskquota are placed in active_tables_map and with a limit of 2
tables overflow occurs even when the extension is created. Secondly, now a
table with 10 partitions is created to overflow active_tables_map, after which
another table is created into which data is inserted that should exhaust the
quota, but since this table does not inserted into active_tables_map, its size
is not taken into account and we can insert into the table after that. At the
end, vacuum full is done to achieve the overflow of altered_reloid_cache.

[KnightMurloc]

we don't need them. I will remake all the tables in shared memory in the same style.