The only dependency the low-level crypto modules have on code in
crypto/obj is their use of OBJ_nid2sn, which is trivial to avoid.
This facilitates future simplification of crypto/obj, including
possibly the removal of functions like OBJ_nid2sn and the complex
build infrastructure that supports them.

This change also removes EVP_CIPHER_name and EVP_MD_name.

Change-Id: I34ce7dc7e58d5c08b52f95d25eba3963590cf2f7
Reviewed-on: https://boringssl-review.googlesource.com/3932
Reviewed-by: Adam Langley <agl@google.com>

Change-Id: Ibc8c3d8740cbd03dfb0d1ff9834fed66601dd479
Reviewed-on: https://boringssl-review.googlesource.com/4123
Reviewed-by: Adam Langley <agl@google.com>

Use it in ssl3_cert_verify_hash so signing a pre-TLS-1.2 handshake hash can go
through RSA_sign and be intercepted via RSA_METHOD appropriately. This avoids
Windows needing to intercept sign_raw. (CAPI keys cannot provide sign_raw,
unless the input size happens to be that of NID_md5_sha1.)

Also use it in processing ServerKeyExchange to avoid special-casing RSA.

BUG=crbug.com/437023

Change-Id: Ia07433f468b75fdf7bfc8fa90c9751639b2478e6
Reviewed-on: https://boringssl-review.googlesource.com/2420
Reviewed-by: David Benjamin <davidben@google.com>

Remove the existing md5_test and sha1_test. They now are all covered by
digest_test. For good measure, test the one-shot functions too.

Change-Id: I8e144cc563fb8817144e26cbd2e10c15642464ba
Reviewed-on: https://boringssl-review.googlesource.com/2211
Reviewed-by: Adam Langley <agl@google.com>