Protect GO_EVP_CIPHER_CTX_PTR with sync.Mutex

[lyoung-confluent]

See golang-fips/go#187 for details on why this is necessary. Protecting these with a mutex is a brute-force approach and presumably has negative performance implications, another option could be to go back to creating a GO_EVP_CIPHER_CTX_PTR within each encrypt/decrypt operation instead of caching the context.

[derekparker]

Thanks for sending this fix!

There's definitely going to be a performance hit when trying to run these operations in parallel. My gut instinct is instead of using this mutex we instead create a new EVP_CIPHER_CTX for each operation instead of reusing, via newCipherCtx and don't store it in any of the structs such as evpCipher. This has the drawback that the caller will have to free that context to avoid leaks, but that's ok. We can also wrap the EVP_CIPHER_CTX in a struct with a finalizer to free the memory when that object is GC'd in the case of a programmer error where free is not called and it starts to leak.

cc @dagood @qmuntal @ueno @simo5

[derekparker]

Closing in favor of #146.