Evaluate replacing JsonConvert serialization

It looks like you're just using JSON.net to serialize `IDictionary` in savedState and this could be accomplished with built-in serializers that would remove the dependency on JSON.net (so that a different version can be used or just not bring that dependency along with someone using System.Configuration.Install).

I looked at the reference source and the built-in classes use `NetDataContractSerializer` which can serialize/deserialize arbitrary types in an `IDictionary`. We don't have that in `netstandard2.0`. Another possibility is `BinaryFormatter`, but that requires `ISerializable` which `IDictionary` is not. The concrete type is `Hashtable` in the underlying implementation for installers.

@flamencist thought [this approach](https://blogs.msdn.microsoft.com/adam/2010/09/10/how-to-serialize-a-dictionary-or-hashtable-in-c/) might work, but it relies on `ToString()` to put it in a key value store. Since `IDictionary` (and `Hashtable`) can hold any object, this doesn't really work. Some objects may not have a sensible `ToString()` which would end up with the type name in the output, and they can't really be deserialized in this manner.

In the `Installer` source I looked at, it appeared that it was putting an integer in there for the count of `Installer`s and then saving all the child `Installer` states (which can be any object since anyone can inherit `Installer`).

The reason `JsonConvert` works is that it serializes the type information too. Is it possible to replace it with something we have in `netstandard2.0` so that we don't have to bring along a JSON.Net dependency?

Start by tracing how Installer savedState is serialized and deserialized, then compare the netstandard2.0 serializer options discussed in the issue, including NetDataContractSerializer and BinaryFormatter. Verify whether arbitrary IDictionary and child Installer state can round-trip without Json.NET; done means identifying a compatible replacement or documenting why the dependency must remain.

由索引模型根據 Issue 內容生成。