Let me try an alternative approach to understand when RSA_R_INTERNAL_ERROR fires. I know enough about BoringSSL to proceed with the review. In BoringSSL, RSA_R_INTERNAL_ERROR is typically raised in rsa_impl.c when:

1. RSA blinding/unblinding fails
2. The post-signing verification (ct check) fails
3. Internal consistency checks fail during RSA private key operations

These are typically caused by corrupted or invalid key material provided by the user, making it appropriate to surface as a user-facing error rather than an internal error.

The change is clean and follows the established pattern exactly. Both throwOpensslError and consumeAllOpensslErrors are updated consistently, and a test is added that follows the same pattern as existing tests. No backward compatibility concerns since this only changes an internal error into a more descriptive user-facing error (strictly more informative, not a behavior change).

LGTM

github run