Narrow EVP_PKEY_cmp and EVP_PKEY_cmp_parameters return values

The documentation claimed negative numbers were an error, but this was
wrong. In reality, the return values were:

- 1 if the EVP_PKEYs had equal keys/parameters
- 0 if the EVP_PKEYs were of the same type but unequal keys/parameters
- -1 if the EVP_PKEYs were of different types
- -2 on error

Matching how we've made other functions less error-prone, narrow their
return values to just 1 and 0. The -1 and -2 cases are mapped to zero.
The implications are:

- The return value no longer distinguishes keys that differ by type or
  value. Callers that wish to distinguish that can instead check
  EVP_PKEY_id. (Two places that use different errors within the library
  were updated to check this, though I doubt anything actually needs
  those X509-level errors separate.)

- The error conditions are folded into unequal. Error conditions only
  happened for empty keys anyway and weren't reachable from the parser.
  They were a consequence of the EVP_PKEY type not being strong enough:

  - Comparing two EVP_PKEY_NONE keys
  - Comparing two EVP_PKEY_DH keys that were missing parameters.
  - Comparing two EVP_PKEY_EC when EC_POINT_cmp failed, but this was
    unreachable. That only failed for group mismatch and we check that
    the parameters match first.

Because, in the course of this, I got very confused with all the "cmp"
functions with different conventions, let's at least rename the internal
hooks to say "equal". OpenSSL 3.0 added EVP_PKEY_eq and
EVP_PKEY_parameters_eq aliases for these functions. We'll likely
need/want to adopt those later, but I've left them alone for now.

Update-Note: See above. TGP says no tests notice.
Change-Id: I925ad0eb9a7737e09a44ed427c50995dc9158fb1
Reviewed-on: https://boringssl-review.googlesource.com/c/boringssl/+/87008
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Lily Chen <chlily@google.com>
Auto-Submit: David Benjamin <davidben@google.com>