"full file" refers to loading the file in its entirety from the disc without parsing it first - and then parsing it from memory.
Decoded mode obviously decodes the file as its reading it, so the "full file" as-it-was on the disc is never entirely in memory - the decoded audio is.
The full file "not decoded" mode will always load the entire file first, close the original, and only then start playing, so it can avoid any complications of switching between different locations of reading the file from etc. This memory is only free'ed after the entire file is done playing.
The "decoded" modes will start decoding audio until either the file is done or the memory is full (or our usage limit exceeded), in which case it'll wait until some audio has played out, which gets free'ed, and then load more. If the file is short enough, or there is enough memory, the entire audio of the file might be in memory at one time.
The difference in behavior is from the potentially vast size difference between encoded and decoded audio. But also from the architecture of how playback simply works.
"Full file (not decoded)" mode is IMHO a much better option to use. It avoids disc access and any problems from that, and uses much less memory.