[llvm] r211668 - Fix another asserting method in the null streamer.

[Rafael Espíndola]

>> Why? We wouldn't want it to assert on the RecordStreamer for example.
>
> This comes down to a question of what the best default is for the base
> class.  If you're implementing a new derived class, is calling this function
> without overriding it a source of error?  If so, the check is worth keeping.
> Otherwise, it isn't.  I don't know the semantics of the MC hierarchy enough
> to address this.
>
> I brought the issue up since I saw three changes in a row (two from you, one
> from Nakamura) which seem to solve similar problems with different
> solutions.  That raised a question in my mind, so I asked.

So, there are only 3 file formats and a handful of streamers.  IMHO
the best design is to move what we can to the target streamers (ARM
being the current worst offender) and what remains in the MCStreamer
interface should be implement in whatever way is most convenient for
the existing streamers. We support plugging in targets, not file
formats.

Given that, it seems that a nop implementation is probably the best.
If the interface gets too cluttered the way forward would probably be
to add a ObjectFormat streamer, similar to what we did for the
targets, but IMHO we are not there yet.

Cheers,
Rafael