Use swift enums instead of raw values

[PaulTaykalo]

Since we have own dictionary, no need to have calculated properties for some parameters in dictionaries.
A lot of time was spent on conversion from stringKind of elements to some specific enum (i.e. SwiftDeclarationKind, StatementKind, SwiftExpressionKind)

These were pre-cached in the dictionary, which allowed to

1. Write a bit more type-safe code
2. Have a bit faster code (there's no need to converts Strings to enums all the time)
3. Some dictionary subscript unneeded access removed as well.

Note

• Some parts of code still using kind as a string because there are no enums or enum cases that represent those strings
• If we're about to continue to use props caching in SourceKittenDictionary, we need to be careful, because if properties will take too much place in memory, SourceKittenDictionary will start behaving like a class and will be put on the heap.

[SwiftLintBot]

	1 Warning
⚠️	This PR may need tests.

	12 Messages
📖	Linting Aerial with this PR took 1.98s vs 2.1s on master (5% faster)
📖	Linting Alamofire with this PR took 3.22s vs 3.54s on master (9% faster)
📖	Linting Firefox with this PR took 11.05s vs 11.71s on master (5% faster)
📖	Linting Kickstarter with this PR took 26.75s vs 27.91s on master (4% faster)
📖	Linting Moya with this PR took 1.97s vs 2.08s on master (5% faster)
📖	Linting Nimble with this PR took 2.45s vs 2.57s on master (4% faster)
📖	Linting Quick with this PR took 1.04s vs 1.07s on master (2% faster)
📖	Linting Realm with this PR took 3.34s vs 3.5s on master (4% faster)
📖	Linting SourceKitten with this PR took 1.62s vs 1.71s on master (5% faster)
📖	Linting Sourcery with this PR took 3.87s vs 4.15s on master (6% faster)
📖	Linting Swift with this PR took 17.81s vs 19.07s on master (6% faster)
📖	Linting WordPress with this PR took 29.93s vs 31.06s on master (3% faster)

Generated by 🚫 Danger

[jpsim]

👍 worth calling out the performance improvements in the changelog too!