While it's true in a mathematical sense that one may not be able to prove that a sequence of seemingly random bits is an encrypted files that is not true in a legal sense. The law isn't about proving things 100%. The law is about weighing the available evidence and proving things to various standards (preponderance of the evidence, beyond reasonable doubt, etc).

For example if if there are server logs showing I downloaded illegal files, and there are people who testify that I talked about downloading illegal files, and there is non pre-installed software on my computer that is used for encryption then probably that sequence of random bits is an encrypted file.

Maybe you can't prove it mathematically, but you can prove it legally.

the standard is "beyond reasonable doubt".

A doubt which you cannot reason away is reasonable, by definition. Therefore it is below the threshold of legal proof.

That's not what the legal term "beyond reasonable doubt" means. Beyond reasonable doubt means that a reasonable person would have no doubt that the party in question was guilty.

An encrypted drive will typically contain metadata identifying it as such, so no.

There is something, I believe called shadow volumes, which are completely metadata less encrypted containers living in a sea of random numbers. While you could claim that the existence of a program able to acess such a volume would be equivalent to metadata, no actual metadata needs to exist, and the random sea could contain one or ten volumes, which without a password you could never know, only guess or assume.

It is could even be possible, even likely, to create encryption schemes where several different encrypted volumes could share the exact same data blocks using something similar to homomorphic encryption. Which raises and obvious question: If the unlocked drive did not contain the data sought, can we hold someone in contempt after they did what we asked from them simply because we didn't find what we were looking for? Because we truly can't know if there is several encrypted volume in the same space without assumptions about information entropy and inaccessible configuration data.

Thankfully homomorphic encryption is not really practically viable today, so that particularly nasty can of worms is not imminent to solve, but we might be well served to let out decisions be informed by it, as it breaks most assumptions of what can be known, and what can't.