Remove usage of the transitive dependency Optimal

[kamilkowalski]

Relates to spandex-project/spandex#124. Currently spandex_phoenix makes use of Optimal even though it doesn't list it as a dependency - it's only available because spandex requires it, and spandex_phoenix requires spandex. This removes the use of Optimal, as we think it is the way to go for Spandex to be fast.

[kamilkowalski]

I think there's something wrong with Coveralls config - tests are passing in the build.

[GregMefford]

Yep, I fixed the CI issues in #45, and I also added an explicit dependency on optimal in that PR, so that we can merge #45 and release a patch version, then merge this one and release a major version since it's changing the behavior in a way that could potentially matter to someone.

[GregMefford]

This looks good to me, but we should merge it after we ship a release with #45 in it, I think.

[GregMefford]

I merged #45. Could you please rebase this onto the updated master branch so make sure the CI is passing again?

[kamilkowalski]

Still failing on uploading results to Coveralls - maybe there's some configuration that prevents sharing secrets in 3rd party PR builds? In case someone opens a PR that prints out the secret to the console.

[GregMefford]

I have returned after a long hiatus and got this CI issue fixed in a separate PR, so now that this is rebased, I think it should be passing. I'd like to move forward on getting this merged, but now it's been so long that I want to take another look through it and make sure it's the right path forward still. Thanks again for doing this work! ❤️

[GregMefford]

Yeah I feel good about merging this at this point, and I have changed my mind about needing a major version for this - it's extremely unlikely to break anyone's existing code in practice because all it would do is not raise a validation error. Really, it's just allowing them to shoot themselves in the foot where they couldn't previously, in order to remove this dependency and improve performance.