the rum index has worked well for us on roughly 1TB of pdfs. written by postgrespro, same folks who wrote core text search and json indexing. not sure why rum not in core. we have no problems.
https://github.com/postgrespro/rum