RFE: devchk design to link even if files don't build

[Jeff Johnson]

The fact that errors abort a devchk build does provide some incentive to fix
specdb data, but... sometimes certain problems can't easily be resolved without
a lot of trouble, and it might be useful to be able to get partial results from
devchk anyway.

Example: new LSB data is incompatible with headers on certain autobuilders. No
matter how much we adjust data, some things will still fail until a new
autobuilder client is put in place.

Thought: each generated test file contains a single function. So "complex.h"
is tested by function complex_h() in file "complex_h.c". Perhaps we could add
a second pass which takes all the test files that failed to build and replace
them with a stub file containing an empty function call, which is then
compiled. This way, the hdrchk binary could link and run, it just wouldn't
provide relevant results for the stubbed functions. The stubbed functions
ought to generate a suitable error message so we don't forget...

Note we can't just link all the .o files into hdrchk, because hdrchk.c is a
generated file with a list of functions to call, so it would fall over calling
functions that didn't get built. We have to somehow fake the ones that failed
to build if we do this.