MDEV-31535: Add privilege-based fast path for SHOW DATABASES listing

[itzanway]

Summary

For users without global database-listing privileges, avoid a full data-directory
scan when building the database list for SHOW DATABASES and
INFORMATION_SCHEMA.SCHEMATA.

Instead of always calling find_files() in make_db_list(), restricted users
can now use a new helper get_acl_databases_for_user() that builds the list
from in-memory privilege tables (acl_dbs, column_priv_hash).

Problem

make_db_list() currently scans the entire data directory via find_files(),
even for users who only have privileges on a small, known set of databases.
This is unnecessary work or restricted users and does not scale well on
installations with many databases.

Solution

1. Add get_acl_databases_for_user() in sql/sql_acl.cc.

   • Collect exact database names from mysql.db and table/column-level grants.
   • Match grantees the same way as the existing visibility checks:
     • db-level: priv_user, active role, and PUBLIC (mirrors acl_get_all3())
     • table-level: priv_user and active role (mirrors check_grant_db())
   • Use sctx->host / sctx->ip for host matching.

2. Gate the fast path in make_db_list() (sql/sql_show.cc) when:

   • the caller is fill_schema_schemata() (SHOW DATABASES / SCHEMATA only)
   • the user lacks global DB-listing privileges
   • all applicable grants use exact database names (no % or _ wildcards)

3. If any applicable grant contains a wildcard, decline the fast path and fall
   back to the existing find_files() scan.

4. Preserve existing output ordering by applying the same sort used by
   find_files() (Discovered_table_list::sort() / sort_desc() in debug
   builds).

5. On fast-path decline, restore the caller's database list to its entry size
   so partial results are not duplicated by the fallback scan.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[janlindstrom]

Looks very promising but requires some cleanup and additional work so that
user can verify is fast path used or not.

[janlindstrom]

Please no white space changes.

[janlindstrom]

Can you please avoid lambda-functions and use inline functions instead.

[janlindstrom]

Similar conditions in two places, consider using inline function

[janlindstrom]

Sure, but how we could verify that fast path was used and cases where not it was not used?
Maybe need again after select schema_name from information_schema.schemata; additional
explain format=json ...; so that it would show fast path in some field...

Check also if you do select schema_name from information_schema.schemata more than once does result come from query_cache or is same code executed again.

[janlindstrom]

Please use strncmp instead