Decaf specs

[savetheclocktower]

Now the specs are in a normal language.

Opened as draft because it depends on #3; will take out of draft once it lands.

[savetheclocktower]

OK, since this is basically mechanical, I'll probably land this PR soon. Mainly just wanted to ensure CI was green!

[savetheclocktower]

Whoa! Moved a helper function and converted it from an expression (var buildSpellChecker = function () {...}) to a function statement (function buildSpellChecker() {...})… and promptly broke 13 specs.

The specs test everything with both the native spellchecker for each platform and hunspell… but the fact that it used a function expression meant that the [true, false] loop in practice tested only the false condition. Because the old version used var, it wasn't block-scoped; it was hoisted out of the loop. Changing var to let would've had the same effect.

So the specs were passing because they weren't testing hunspell! Luckily, 10 of the 13 failures related to using hunspell were because the spec did not set a language first — which doesn't actually happen as part of our usage of this library in Pulsar, so it was easy to get them to pass once we set a language.

The other three failures were due to poor guards. One spec purposefully tests macOS native spellchecker behavior, but guarded only on process.platform === 'darwin', so could never have passed in the hunspell branch. Now we guard on the specific spell-checking adapter rather than simply the platform. Another spec described “returns nothing for a pair of 257 1-byte character strings with encoding,” but didn't test that; it tested two-byte strings in a way that contradicted its sibling test as well as its own description. Restoring the clear intent fixed the test.

These specs were broken in the CoffeeScript spec file, so they must've been broken for years.

[savetheclocktower]

Oh, also: added some extra specs for the singleton pattern — which previously was not tested because the tests always constructed a new Spellchecker instance.