This also wouldn't really work because a YouTube video isn't a single file. Videos are available in different qualities and formats. The video component tends to be separated from the audio component. If you want to combine them into a single file, you have to mux them. However, this process is not deterministic, so the resulting hash would differ depending on a multitude of factors.
Then Youtube would have to publish the hashes for each and every file that they encode the source material to. Practically impossible and especially unverifiable if YouTube ever disappears (or the source gets deleted).
A digital signature of the hash, embedded in the metadata and computed similarly to Reproducible Builds... doesn't take much space and especially only requires knowledge of the public key for verification.