The s-values produced by secp256k1_surjection_genrand previously
depended only on their indices and the difference between
input_blinding_key and output_blinding_key. Calls with the same
difference therefore reused s-values even when their proof inputs
differed.
For proofs with the same used-input selection and honest input index,
the same generated s-value was used as the signing nonce. Reusing this
nonce across different proof messages allowed recovery of the
blinding-key difference. The remaining repeated s-values also revealed
the honest input index.
This commit affects proof generation only; verification is unchanged.
65093e1444