[LOW] Normalize malformed signature verification errors

[OskarEichler]

Summary

Normalize malformed signature and verification-data inputs across ECDSA, RSA-PKCS1, RSA-PSS, and EdDSA to SignatureVerificationError. This keeps attacker-controlled invalid signature material inside the gem's documented verification-failure boundary instead of leaking Ruby, OpenSSL, or Ed25519 backend exceptions.

The original commit normalized malformed ECDSA byte strings that raise OpenSSL::PKey::PKeyError. The follow-up validates input types before backend calls and normalizes invalid Ed25519 signature lengths.

Reproduction

Release/current behavior can raise NoMethodError, TypeError, ArgumentError, or OpenSSL::PKey::PKeyError for malformed signature shapes and lengths. Downstream callers such as COSE commonly catch OpenSSL::SignatureAlgorithm::Error; the leaked exceptions can therefore turn an invalid credential into an unhandled request or worker failure.

The focused model exercises eight malformed signatures and five malformed verification-data values across all four algorithms: 52 rejection paths in total.

Verification

• Candidate: 52/52 malformed inputs raise SignatureVerificationError; all four algorithms still verify valid signatures
• Existing suite: 31 examples, 0 failures on Ruby 4.0.6 through rbenv
• RuboCop: 20 files, 0 offenses
• Ruby 3.2.11 syntax checks pass for all three changed paths
• Version 1.3.0 gem builds and unpacks successfully
• No repository tests or dependency files changed

Compatibility and breaking changes

Valid signatures, key assignment/deserialization, and signing behavior are unchanged. Invalid non-string inputs and malformed Ed25519 lengths now raise the gem's existing SignatureVerificationError instead of backend-specific exceptions; callers relying on those backend exception classes will observe this intentional error-normalization change.