This is primarily for access to comments, which would be expected to be available in a protoc plugin.

The implementation has two fiddly aspects:

- We use a Lazy<T> to avoid building the map before cross-linking. An alternative would be to crosslink at the end of the constructor, and remove the calls to CrossLink elsewhere. This would be generally better IMO, but deviate from the Java code.
- The casts to IReadOnlyList<DescriptorBase> are unfortunate. They'll always work, because these lists are always ReadOnlyCollection<T> for a descriptor type... but we can't use IList<DescriptorBase> as that's not covariant, and it's annoyingly fiddly to change the field to be of type ReadOnlyCollection<T>.

This consists of:
- Changing the codegen for the fixed set of options protos, to parse unknown fields instead of skipping them
- Add a new CustomOptions type in the C# support library
- Expose CustomOptions properties from the immutable proto wrappers in the support library

Only single-value options are currently supported, and fetching options values requires getting the type right
and knowing the field number. Both of these can be addressed at a later time.

Fixes #2143, at least as a first pass.

Generated code changes for previous commit (basically InternalBuildGeneratedFileFrom => FromGeneratedCode)

There are corner cases where MessageDescriptor.{ClrType,Parser} will return null, and these are now documented. However, normally they *should* be implemented, even for descriptors of for dynamic messages. Ditto FieldDescriptor.Accessor.
We'll still need a fair amount of work to implement dynamic messages, but this change means that the public API will be remain intact.

Additionally, this change starts making use of C# 6 features in the files that it touches. This is far from exhaustive, and later PRs will have more.

Generated code changes coming in the next commit.

Changes in brief:
1. Descriptor is now the entry point for all reflection.
2. IReflectedMessage has gone; there's now a Descriptor property in IMessage, which is explicitly implemented (due to the static property).
3. FieldAccessorTable has gone away
4. IFieldAccessor and OneofFieldAccessor still exist; we *could* put the functionality straight into FieldDescriptor and OneofDescriptor... I'm unsure about that.
5. There's a temporary property MessageDescriptor.FieldAccessorsByFieldNumber to make the test changes small - we probably want this to go away
6. Discovery for delegates is now via attributes applied to properties and the Clear method of a oneof

I'm happy with 1-3.
4 I'm unsure about - feedback welcome.
5 will go away
6 I'm unsure about, both in design and implementation. Should we have a ProtobufMessageAttribute too? Should we find all the relevant attributes in MessageDescriptor and pass them down, to avoid an O(N^2) scenario?

Generated code changes coming in the next commit.

We'll see what I've missed when CI fails...

Update code already within Google.Protobuf.Reflection which was importing Google.Protobuf.DescriptorProtos

- Move types into Google.Protobuf.Reflection
- Change codegen to reflect that in generated types

Generated code changes coming in part 2

- No parsing
- Reflection based, so not hugely efficient
- No line breaks or indentation

- The protos are no longer publicly exposed at all
- Oneof detection now works (as we default to -1, not 0)
- OneofDescriptor exposes the fields in the oneof
- Removed unnecessary code for replacing protos - remnant of extensions
- There's now just the non-generic form of IDescriptor

Everything should now be consistent - the only difference between files is the year (aside from generated files).

Fixes issue #531.