Here is another idea:

Create a companion process which runs alongside the Firebird server. It
gets notified via events when records are updated and looks to see if any
plaintext fields are empty when the RTF field is not empty. If so, it reads
the RTF, converts to plaintext, fills in plaintext field. Along the way it
parses plaintext and indexes it into the words table.

In this way you remove the responsibility for parsing and indexing from the
database server proper but still have a "server" process responsible for
the task so client apps are not burdened with this. The indexing server can
use whatever components are appropriate and not have to deal with the
restrictive UDF operational context.