Fix RSAPublicKey hierarchy and hidden virtual functions

[maxtropets]

List of problems to solve while refactoring the interface for signature verification

• RSAPublicKey and RSAKeyPair both define verify(signature_args..., md_type, salt_length), is it even legit?..
• RSAKeyPair_OpenSSL implements verify(signature_args..., md_type, salt_length), BUT PublicKey_OpenSSL implements verify(signature_args..., md_type, hash_bytes&). Here's why JWT authentication uses the wrong verification impl (check comment).
• salt_length is size_t, therefore it's not possible to pass options like RSA_PSS_SALTLEN_AUTO == -2 or other predefined constants. Not sure what's the best way to do it in the interface, it's TBD how different paddings and salt work as per documentation first, but this has to be fixed.

[maxtropets]

A sketch partially displaying the current state.

I'm confused about a real purpose of having verifying methods in (RSA)KeyPair interface, while we could have KeyPair as a child of PublicKey, so we won't copy verification interface, as well as public_key_der(), public_key_pem, etc.

[maxtropets]

Suspected #6405 to hide the verify(args..., md) overloads by adding salt_size, but ruled that out, because it turns out the PublicKey::verify(args, md) is not virtual (🤦).

Instead, it has an implementation, which computes a hash of the contents and them calls virtual verify(args, md, hash_out)

However, salt_size shall also be removed, as it doesn't fit the common interface. I'm thinking of some sort of separate params struct/variant to pass depending on the underlying key type, and leave signature and contents as the only interface.