This seems to be a larger problem with the KMS implementation. Currently the sign_bytes that are returned in session.rs/sign() change depending on whether the KMS is compiled with the YubiHSM backend or the Ledger backend. This should never be the case since the KMS should be agnostic to the backend being used.

So I've added logging after the sign_bytes method in called in session.rs/sign(). With the YubiHSM it results in
Proposal: 119
Prevote: 109 or 110
Precommit: 109 or 110

With the ledger it results in:
Proposal: 120
Prevote: 37

However the sign() method is called before it ever reaches any specific backend HSM. One reason for these errors could be that due to the conditional compilation with "--features" the KMS generates different data depending on which features are turned on.

@tarcieri @liamsi I have no idea why this happens. With the Yubi HSM it works fine and sign_bytes returns the correct values, whereas if you don't compile with yubihsm the returned sign_bytes are wrong.

The second issue here is that the Ledger validator application does not expect these sign_bytes, since they are divergent from the specification. The reason why this only appears with the ledger is that the YubiHSM will just sign whatever it is given, but the ledger actually decodes the bytes. @jleni