My understanding is that all that is necessary is to just feed the PCM
stream into the fingerprinter, like any other decoder would.
Yes, that's correct. Picard already supports fingerprinting, so once you have the audio data the rest is trivial. A lot of the discussion misses the point. Feeding audio data to the fingerprinter is trivial, doing audio format conversion isn't necessary (and even wouldn't be too hard if it was) and whether the data is stored completely in memory or in a temporary file is just an implementation detail. But you have to extract (trying to avoid the word ripping here as it confuses folks) the audio data from the disc, and that's the messy part. You basically implement 80% or the functionality of a full CD ripping software for this, and have to do so in a cross platform way.
Not saying it is impossible or anything, just that the relatively high effort and the added complexity in Picard makes it unlikely this will happen soon.