It depends on a lot of things really, like how much time do you have to get it running (day, weeks, months, years ?), what is your current experience, and lots more. At this point I would definately choose 19c for a lot of reasons, one that 19c is currently in its peak (long enough to not have lots of bugs, and long enough to have all documents on point, and long enough to the end-of-support). But also, if you run 19c databases, there's no doubt that technically that runs better in a 19c Cluster than in a 26ai Cluster. For example you have to download and install and test more patches in CPU context, if the versions are different, just to even start.

"if you run 19c databases, there's no doubt that technically that runs better in a 19c Cluster than in a 26ai Cluster."

I would disagree with that statement. I've run different RDBMS versions of Oracle on clusterware that does not match the RDBMS version and I've never had any problems. Not only will 19c RDBMS run just fine on 26ai Grid Infrastructure, but this is a certified combination. In fact, you are going to upgrade to 26ai in the future and you will have to upgrade GI first before you can even install the RDBMS software. I typically run GI at a higher level for at least 2 weeks before upgrading the RDBMS software, and usually much longer than 2 weeks. I'm not sure what would run "better". You'd have to elaborate on that point.

To fix patch 38523609 there are 3 files that need to be edited to add the same single line of code to each. Oracle wrote a PL/SQL "select … into …" statement without an exception clause to catch an error where no records were found. (thanks for the help @anilech)

In the zip file it's:
38523609/files/sqlpatch/38523609/28341036/rollback_files/19.1.0.0.0/javavm/install/bug35933540_rollback.sql
38523609/files/javavm/install/bug35933540_rollback.sql
38523609/files/javavm/install/bug35933540_apply.sql

If you have applied the patch, you'll need to edit the files in your ORACLE_HOME:
$ORACLE_HOME/javavm/install/bug35933540_apply.sql
$ORACLE_HOME/javavm/install/bug35933540_rollback.sql
$ORACLE_HOME/sqlpatch/38523609/28341036/rollback_files/19.1.0.0.0/javavm/install/bug35933540_rollback.sql

There is a PL/SQL block that needs to be modified on lines 30-47 (broken code below):
30 declare
31 stat varchar2(30);
32 begin
33
34 select status into stat from dba_registry where comp_id = 'JAVAVM';
35 if stat != 'REMOVED' then
36 begin
37 execute immediate 'revoke select on SYS.DBA_JAVA_CLASSES from SELECT_CATALOG_ROLE';
38 execute immediate 'revoke select on SYS.CDB_JAVA_CLASSES from SELECT_CATALOG_ROLE';
39 execute immediate 'grant read on SYS.DBA_JAVA_CLASSES to public with grant option';
40 execute immediate 'grant read on SYS.CDB_JAVA_CLASSES to public with grant option';
41 exception
42 when others then
43 if sqlcode not in (-1927) then raise; end if;
44 end;
45 end if;
46 end;
47 /

The fixed code:
30 declare
31 stat varchar2(30);
32 begin
33
34
select status into stat from dba_registry where comp_id = 'JAVAVM';
35 if stat != 'REMOVED' then
36 begin
37 execute immediate 'revoke select on SYS.DBA_JAVA_CLASSES from SELECT_CATALOG_ROLE';
38 execute immediate 'revoke select on SYS.CDB_JAVA_CLASSES from SELECT_CATALOG_ROLE';
39 execute immediate 'grant read on SYS.DBA_JAVA_CLASSES to public with grant option';
40 execute immediate 'grant read on SYS.CDB_JAVA_CLASSES to public with grant option';
41 exception
42 when others then
43 if sqlcode not in (-1927) then raise; end if;
44 end;
45 end if;
46 exception when no_data_found then null;
47 end;
48 /