Even if an add-on advertises itself as compatible, it does not mean the actual add-on code was edited to use changes from NVDA itself. This is more so if a version of NVDA introduced changes that can break certain add-ons, most notably speech synthesizer add-ons, made complicated if something (a different add-on, perhaps) other than these made internal changes to (patched) NVDA that renders some synthesizers inoperable or fail to load properly.
As for add-ons and installing NVDA from portable versions: yes if the user checked "copy portable user configuration" checkbox. If one or more incompatible add-ons are detected, users might be asked to review them.
As for running NVDA with or without speech: yes, it is possible to run NVDA without speech. Speaking of speech syntehsizers, if an external speech synthesizer fails to load, NVDA will resort to either eSpeak NG or Windows OneCore voices if running on Windows 7/8.x or Windows 10 and later, respectively.
Regarding TripleTalk issue, the log reports an issue with add-on manifest validation. This is because the add-on summary data is not formatted as a proper Python string. Without going into technical details (ask on NVDA add-ons list or me privately), if the add-on manifest data is not written correctly, Python (NOT NVDA) will recognize manifest data as something other than text. More specifically: