Ah, i see so it is more theoretical
OK, so you seem to mix up uncompressed PCM Audio with the lossy compressed Codec. Let me try to explain.
Again, AAC is a compressed codec. So the encoded bitstream as it resides on harddrive or as it is transferred via stream does not have a fixed size of Bits per sample. Only uncompressed Audio formats have a fixed number of bits per sample.
AAC will compress whatever you input as good as it can to the target bitrate, so it does not matter at all if the uncompressed input was 16 bit or 32 bit. In the end, the outgoing bitrate needs AAC to cut off this and that parts of the wave curves anyway.
But, for explaination: of course, BEFORE the audio is encoded it WAS uncompressed and also it is uncompressed again directly after decoding (right before it goes from RAM to the Sound Card in a player like VLC). BUT it does not matter for the AAC Codec which format it was before, and as it cannot be restored Lossless, it also does not matter which format it is after Decoding.
In the end, most lossy audio compressors only support a single input bit resolution because and just does not matter which one they choose, they just go for the best one.
Why choosing 32bit float for input and output from and to a lossly audio encoder: because literally any other format (<=32bit) can be converted into 32bit float 100% lossless.