Improve option normalization

[szmarczak]

This allows us to normalize options on-the-fly. This will replace got.normalizeOptions / Request.normalizeOptions. The following should give a significant performance boost when called

import {Options} from 'got';
const options = new Options(); // or new Options({url: 'https://localhost'})

options.url = 'https://localhost';
options.url; //=> URL

options.url = 'invalid'; //=> throws

(async () => {
	for (let i = 0; i < 1000; i++) {
		await got(options);
	}
})();

Also it will be a lot easier to test the options.

TODO

• Make some properties non-enumerable
• Normalize Promise API arguments
• Implement option merging
• Implement defaults
• Fix Promise API
• Fix Got interface
• Fix benchmarks
• Invert option.methodRewriting mechanism (fixes Invert the methodRewriting option logic #1307)
• Fix Got options -> HTTP(S) options Proxy handler. Missing has() and ownKeys().
• ...

Checklist

• I have read the documentation.
• I have included a pull request description of my changes.
• I have included some tests.
• If it's a new feature, I have included documentation updates in both the README and the types.

[szmarczak]

Of course it still will be possible to call got(url, {headers: {...}, ...}) or just got({headers: {...}, ...}).

[szmarczak]

Also it will throw when passing a non-existing property on the Options. Migrate:

-got({token: 'asdf'});
+got({context: {token: 'asdf'}});

[szmarczak]

It gives a significant performance boost in terms of normalization:

got - normalize options x 95,800 ops/sec ±2.19% (91 runs sampled)
got - new options x 202,692 ops/sec ±0.33% (96 runs sampled)

It's 2.1x as fast as got.normalizeOptions!

[szmarczak]

Ok, so console.log(new Options()) also logged internals, which would be OK as the getters aren't logged by default. So I wanted to make this[INTERNALS].body, this[INTERNALS].json and this[INTERNALS].form not enumerable and the performance dropped back to 95k.

So instead I did make the entire INTERNALS not enumerable and it "kind of" solved the problem because the benchmarks now look like this:

got - normalize options x 95,929 ops/sec ±1.81% (94 runs sampled)
got - new options x 182,247 ops/sec ±0.72% (98 runs sampled)

1.9x is also great. But I wanted all the getters to get logged by default, but encountered this.

[szmarczak]

Also the Options template now requires to pass an ErrorType. This is to extract the types to a different file. In the end Got will export Options = Options<RequestError> anyway, so the types remain the same.

[szmarczak]

Ok, I've got the streams working, which is pretty nice!

https://github.com/szmarczak/got/blob/8798afc0c5f3ac4bbecde9e49c83cb2cb33722d6/demo.js

[sindresorhus]

Is this a breaking change?

[szmarczak]

probably yes but the final say should be when I finish this PR. I don't expect any breaking changes than option normalization (but it should still be 99% compatible for the end user). I'm short on time, preparing for uni exams hence I'm not so active since last week :(

[szmarczak]

Closing in favor of #1667