Restructuring of PdfViewer code.

The only change in behavior is that SkPdfAllocator on SkPdfContext is
no longer allocated on the heap.

In general, I have just moved code.

Inherit from SkNoncopyable.
Make SkPdfContext directly own fTmpPageAllocator. fTmpPageAllocator is
created when SkPdfContext is, and destroyed at the same time as well,
so there is no reason for the extra allocation.
Add the function parseStream. This eliminates code duplication, and
allows making fTmpPageAllocator private.
Move PdfMainLooper into the implementation file, since it is now only
used by parseStream.
Move SkTDictWithDefaultConstructor and render stats info here, in
support of PdfMainLooper.

Rename PdfTokenLooper to SkPdfTokenLooper.
Move readToken here, unchanged.
Remove tokenizer(), which is unused.

Remove tokenizerOfPage and tokenizerOfBuffer, which are unused.

Move gPdfOps and PdfOperatorRenderer into a header file (hidden for
now), so they can be accessed by both SkPdfRenderer.cpp and

Harvest things into other files:
PdfMainLooper (and the code that calls it) -> SkPdfContext.
readToken -> SkPdfTokenLooper.

Review URL:

git-svn-id: 2bbb7eff-a529-9590-31e7-b0007b416f81
diff --git a/pdfviewer_lib.gyp b/pdfviewer_lib.gyp
index 37b1285..405d02c 100644
--- a/pdfviewer_lib.gyp
+++ b/pdfviewer_lib.gyp
@@ -18,6 +18,7 @@
+        '../experimental/PdfViewer/src/SkPdfTokenLooper.cpp',