[PORT] cache Antlr parse result

[tomlm]

Port this change from botbuilder-dotnet/master branch:
microsoft/botbuilder-dotnet#3621

The import chain of LG file is like a graph, there may be various reference logic.
Antlr's parse takes a lot of time in it, If LG SDK can cache the results of Antlr parse(single call life cycle), the performance will be greatly improved.

After experiments, it can save more than half of time for complex lg call chain.

issue link: microsoft/BotFramework-Composer#2374

Changed projects

• Microsoft.Bot.Builder.LanguageGeneration