Framework: Optimize block validation attribute testing

[aduth]

Related: #5897

This pull request seeks to optimize the block validator, specifically in its handling of attributes, replacing use of Array with Set, where lookup via Set#has is many magnitudes more performant than Array#indexOf (~4x faster with CoreJS shim, ~9x faster natively). To my surprise while testing, I was not aware that Set is supported in IE11. Our build process still applies CoreJS, though I'm unsure if it is applied as a proper shim (when not already present in environment). In any case, a further step may be to drop this, though this should probably already be handled via babel-preset-env.

Testing instructions:

Verify that tests pass, and there are no regressions in the behavior of block validation (including invalidation), particularly affecting "meaningful attributes" (boolean attributes like disabled or enumerated attributes like contenteditable="true").

[aduth]

Actually, upon a more fair consideration (smaller set size), the difference is not as noticeable. A native Set#has on a set of 10 items is still faster than Array#indexOf by about 15%, but the shimmed CoreJS Set#has is actually half as fast as Array#indexOf on the same small set of items. (jsperf)

[aduth]

Unfortunately, even in browsers where Set is supported, we're apparently still forcing the polyfill:

MEANINGFUL_ATTRIBUTES.constructor === __WEBPACK_IMPORTED_MODULE_3_babel_runtime_core_js_set__.default
// true

I think moving forward with these changes would only be worthwhile if this could be resolved.

[mcsf]

I think moving forward with these changes would only be worthwhile if this could be resolved.

Agreed. Set is definitely a good step, but should we close this until then?

[aduth]

Sure, let's. In the meantime, I think I'll keep Set in #5897 as an anticipated future optimization.