[Annotated Buffers] Improve efficiency

[dbaileychess]

Changed the bulk of the text-based annnoted buffers to use a std::ofstream instead of building a giant std::string in memory and flushing the whole thing. It was easy to do since we build in file order.

Added --annotate-sparse-vector option which, when set, will exclude all but the first/last vector element:

Example:

vector (AnnotatedBinary.Foo.foobars):
  +0x00D4 | 03 00 00 00             | uint32_t   | 0x00000003 (3)                | length of vector (# items)
  +0x00D8 | 34 00 00 00             | UOffset32  | 0x00000034 (52) Loc: 0x010C   | offset to union[0] (`Bar`)
  <2 regions omitted>
  +0x00E0 | 04 00 00 00             | UOffset32  | 0x00000004 (4) Loc: 0x00E4    | offset to union[2] (`Bar`)

This will be useful when annotating giant files that are just vectors, where the contents don't really matter to the end user.