Fix memory leak if ossl_bn_new() fails

[ndossche]

When that call fails, the bn BIGNUM is never freed in asn1integer_to_num(). To solve this, use rb_protect().

Example Valgrind report:

32 (24 direct, 8 indirect) bytes in 1 blocks are definitely lost in loss record 11,113 of 25,910
  malloc (at /usr/libexec/valgrind/vgpreload_memcheck-amd64-linux.so)
  CRYPTO_zalloc (at /usr/lib/x86_64-linux-gnu/libcrypto.so.3)
  BN_new (at /usr/lib/x86_64-linux-gnu/libcrypto.so.3)
  BN_bin2bn (at /usr/lib/x86_64-linux-gnu/libcrypto.so.3)
  <unknown stack frame>
 *asn1integer_to_num (ossl_asn1.c:136)
 *asn1integer_to_num_i (ossl_asn1.c:165)
  rb_protect (at /usr/lib/x86_64-linux-gnu/libruby-3.2.so.3.2.3)
 *decode_int (ossl_asn1.c:356)
 *int_ossl_asn1_decode0_prim (ossl_asn1.c:777)
 *ossl_asn1_decode0 (ossl_asn1.c:936)
 *ossl_asn1_decode_all (ossl_asn1.c:1058)
  <unknown stack frame>
  <unknown stack frame>
  <unknown stack frame>
  rb_vm_exec (at /usr/lib/x86_64-linux-gnu/libruby-3.2.so.3.2.3)
  <unknown stack frame>
  <unknown stack frame>
  rb_catch_obj (at /usr/lib/x86_64-linux-gnu/libruby-3.2.so.3.2.3)
  <unknown stack frame>
  <unknown stack frame>
  <unknown stack frame>
  rb_vm_exec (at /usr/lib/x86_64-linux-gnu/libruby-3.2.so.3.2.3)
  rb_yield (at /usr/lib/x86_64-linux-gnu/libruby-3.2.so.3.2.3)
  rb_ary_each (at /usr/lib/x86_64-linux-gnu/libruby-3.2.so.3.2.3)
  <unknown stack frame>
  <unknown stack frame>
  <unknown stack frame>
  rb_vm_exec (at /usr/lib/x86_64-linux-gnu/libruby-3.2.so.3.2.3)
  rb_yield (at /usr/lib/x86_64-linux-gnu/libruby-3.2.so.3.2.3)
  rb_ary_each (at /usr/lib/x86_64-linux-gnu/libruby-3.2.so.3.2.3)
  <unknown stack frame>
  <unknown stack frame>
  <unknown stack frame>
  rb_vm_exec (at /usr/lib/x86_64-linux-gnu/libruby-3.2.so.3.2.3)
  rb_yield (at /usr/lib/x86_64-linux-gnu/libruby-3.2.so.3.2.3)
  rb_ary_each (at /usr/lib/x86_64-linux-gnu/libruby-3.2.so.3.2.3)
  <unknown stack frame>
  <unknown stack frame>
  <unknown stack frame>
  rb_vm_exec (at /usr/lib/x86_64-linux-gnu/libruby-3.2.so.3.2.3)
  <unknown stack frame>
  <unknown stack frame>
  rb_catch_obj (at /usr/lib/x86_64-linux-gnu/libruby-3.2.so.3.2.3)
  <unknown stack frame>
  <unknown stack frame>
  <unknown stack frame>
  rb_vm_exec (at /usr/lib/x86_64-linux-gnu/libruby-3.2.so.3.2.3)
  rb_vm_invoke_proc (at /usr/lib/x86_64-linux-gnu/libruby-3.2.so.3.2.3)
  rb_proc_call_kw (at /usr/lib/x86_64-linux-gnu/libruby-3.2.so.3.2.3)

This was found by a hybrid static-dynamic analyser that looks for inconsistent handling of error checks in bindings.

[rhenium]

I think a cleaner solution may be to prepare a wrapped BIGNUM first and then let ASN1_*_to_BN() write into it directly. ossl_pkey_ec.c does this in a few places.