proof file: optimize appending proof

[guggero]

Background

Additional context: #419 (comment)

The current implementation of the proof file is not very efficient when it comes to appending a new proof: The whole file is read into memory, the new proof is encoded and then the whole file is written again.

This can be optimized in the following way:

• read the number of proofs from the beginning of the file
• skip to the current last proof in the file, only read that proof and hash it
• append the new proof and update the incremental hash, using the previous proof's hash
• seek back to the beginning of the file (depends on what type of io.Reader we actually have) to update the count of the number of proofs

[guggero]

@ben2077 is going to take a look at this.

[Roasbeef]

Related to #147