Fixes a couple of bugs on windows caught by AppVerifier

• in function mz_crypt_pbkdf2(..), CryptDeleteHash was being called after the associated Crypt context was released
  • hmac1 and hmac2 are created using mz_crypt_hmac_init, so they each get their own context (provider) from the CryptAcquireContext(..) function
  • hmac3 is created using mz_crypt_hmac_copy, which creates a duplicate of hmac2's hash, sharing the same context
  • During destruction, hmac1 and hmac2 were being deleted first, followed by hmac3. This meant that the crypt contexts were released first by CryptReleaseContext. The subsequent call to CryptDestroyHash(..) on hmac3's hash would lead to UB as the associated context has already been released by that point
  • Fix: call mz_crypt_hmac_delete(..) on hmac3 first, so that the contexts are cleared at the end after the hashes are destroyed.
• in function mz_crypt_hmac_copy, if target->hash already contained a hash prior to the function being called, that hash would be overwritten, without CryptDestroyHash being called on it.
  • Fix: check if target->hash is non-empty in mz_crypt_hmac_copy and call CryptDestroyHash on it