Move context.IsPreProcessorKeywordContext to beginning of all checks in overrides of AbstractSyntacticSingleKeywordRecommender.IsValidContext.

Remove check for 'IgnoreCase In CSharpAddImportCodeFixProvider we
explicitly override IgnoreCase to be false.  Therefore, we will never use
a case in-sesitive comparer.

1) Reduce number allocations in the analyzer driver that show up perf traces for Picasso.
2) Currently we maintain a cache of CompilationWithAnalyzers for projects being actively analyzed. This is causing the VM to rise and also forcing GC to kick in at different phases causing regressions at various scenarios. The fix is to just create a new CompilationCompilationWithAnalyzers instance for each document/project analysis request.

speculative semantic model determine if we should use the fully qualified
name or not. What this approach failed to account for was if the name
was going to be simplified further due to the containing namespace. To
Fix this we run the simplifier over our document and check if there are
any changes made to the namespace. If this is the case, then we re-sort
the usings according to the simplified name.

Additionally, the IsDebugger and IsImmediateWindow properties have been removed from CompletionTriggerInfo. Any code that was checking these properties now checks for options that are set by the Controller in debugger scenarios.

Fix #732. C# completion list must not suggest `public` as a potential
accessibility modifier to use inside an `interface { }` block;
otherwise, it produces code that does not compile.

Before this change, showing MAS for a type with 4000 empty methods took about 2
minutes on my machine. With this change, it now takes 2-4 seconds. There were
2 different issues causing the slow generation:

1. When formatting the generated file, the formatting engine needs to find the
distance between every token and the new line (or start of file)  before it.
This calls  TokenStream.GetOriginalColumn, which computes the original column
by walking backwards through its list of tokens counting the width of tokens
and trivia until it finds a newline. Though this n ^ 2 operation takes place
in all invocations of the  formatter, it is normally ameliorated by the fact
that real code has enough newlines that GetOriginalColumn doesn't actually
traverse back to the first token in the file. Generated files have no newlines,
so GetOriginalColumn takes up about half the time of generating a MAS document,
because GetOriginalColumn is called on every token in the file.
To mitigate this, we augment MAS' custom formatting rule to not calculate
anchor indentation. Anchor indentation is what requires the original columns,
but no operations are produced for MAS documents, which have no newlines
in them.

2. If instructed to use "auto insertion location", the code generation service
computes the location to insert each member, one at a time. This computation
checks that the type's members are sorted, resulting in n ^ 2 iterations over
the list of members. The fix for this takes Kayle's fix for enum members a
little further:
When instructed not to use "auto insertion location", the code generation
generates declaration syntax for each member without inserting it. It then
sorts that list of declarations once, and bulks adds them (essentially
appending each declaration to the end of the growing type declaration).

I've fixed up some of the "Create*Declaration" methods to match the behavior
of the corresponding "Add*Declaration" methods.

This change does two actions to our public API files:

- Resets them to the versions from 1.0.0 RTM
- Splits them into the shipped and unshipped files

Add test for Block Context within AutoProperty Initializer. Add better explaining comments about preferring block over initializers. Addresses code review feedbacks.

Fixes #936. This allows introduce variable codefix to function in autoproperty initializers for both VB and C#.