create new branch, e.g jira/solr-11701
mvn dependency:tree on the newly released Apache Tika and MEMORIZE it
upgrade all dependencies in lucene/ivy-versions.properties – make sure that they are in alphabetical order
add any new licenses in solr/licenses – must include a -LICENSE-XYZ.txt and -NOTICE.txt file for every jar
update anything new in solr/contrib/extraction/ivy.xml
ant clean (out of nervous habit) and then run the unit tests in contrib/dataimporthandler-extras and contrib/extraction
Fix any problems in the source code, and this can include XLSXResponseWriter which relies on Apache POI.
ant clean-jars jar-checksums
git add new .sha1 files in solr/licenses and lucene/licenses and git rm old .sha1 files
ant precommit
Receive immediate errors that you missed something and go back two steps; repeat ant precommit as needed, waiting 15-20 minutes each time ... if you didn't break something obvious.
In my environment, ant precommit eventually ends in errors about broken links in html. This means you are successful!!!
Run ant test for kicks. Something will likely break. Try to figure out if it is caused by anything you did or just a flaky build. Bonus points if the test failure is reproducible and you report it/fix it.

Phase 2: Integration Testing Solr

...

Phase 4: Reflect, Rejoice, Work

Reflect on:
1. The tedium to get the dependencies right and the risks of not getting them right
2. The ever present risks of jar hell by integrating Tika into Solr
3. The seductive belief that Tika won't break Solr, when we know it will eventually, and we should really be keeping Tika out of Solr if at all possible...and yet maintain the awesome easy-to-get-started-ness of the current integration.
Work on~~SOLR-7632 and/or SOLR-11721~~ TIKA-3093!
Rejoice that Tika is being refactored out of Solr in 9x.
Work towards whatever solution allows for an easy, out of the box extraction process for binary files