"Source code" is generally code you can put into a compiler and get out object code. The term "source" refers to the fact that it is used as input. This is why, for example, the GPL specifies that you must release the original source code in the preferred format for editing — so that people wouldn't release source code generated by disassembly or a source-to-source transformer.
I suppose it all depends on how you define "source code". For me, source code implies that it was authored by a human or automatically generated to mimic human authorship. Any automated translation step that does not result in executable "object code" or "machine language" is "intermediate code".
It may be usable in the same way as source code, but since it was produced as the output of a program that had an input closer to the source, it cannot be the source.
I imagine the compilation process like a stream. The source is the origin of the flow. All changes made there propagate downstream. Some streams are short, like those with M4-processed assembly. Others are longer, with MSIL or JVM intermediate code. Linked libraries are like tributaries; they have their own sources. The end product is a single river, which fans out into a delta for each supported processor architecture as it nears the sea of end users.
