[DynInst_API:] commit: assorted fixes, pre-9.0.4

[Bill Williams <bill@xxxxxxxxxxx>]

Got a whole stack of these that have been pending:

* Fixed bug in per-CU line information parsing that prevented line info 
from being parsed in certain cases.
* Shared libraries are now consistent with executables: .o files are 
modules/mapped_modules, and they are contained in objects/mapped_objects.
* Fixed CMake exports to match what the test suite expects.
* Re-enable building 32-bit runtime library, but default to "no".
* Documentation fixes and new code sample (disassembler) (patches from 
Alin Mindroc)
* Fixed intermodule variable references.
* Fixed performance problem in line information parsing where we'd 
repeatedly parse a single CU's information and replicate that data each 
time.
* Additional boost versions added to packages.cmake (through 1.59).
* Fixed Windows performance problem: relocation/instrumentation could 
trigger unnecessary parsing
* Fixed unsafe string operations
* Fixed LD_PRELOAD logic (patch from Victor van der Veen)
* Fixed isDynamic crash (patches from Victor van der Veen and Marc Bruenink)

Known issues:

* Build failure on certain Cray platforms (and possibly others?) with 
numeric limits macros somehow not being properly defined. My current 
thought is to C++-ify everything and redefine the macros in Types.h to 
the appropriate std::numeric_limits call, which should(?) be more 
universally portable these days (and requires less preprocessor voodoo). 
Thoughts from the list?
* Cross-compilation on Intel MIC systems may require extra boost 
definitions in order to build with the back-end (k10m) compiler.
If you're using BPatch_module, BPatch_object, Symtab::Module, or 
Symtab::Object on shared libraries, *please* test this out and let me 
know what interfaces on the object classes should be there and aren't 
(and whether this change is a horrible backwards compatibility breaker).
Enjoy; I hope to be able to make 9.0.4 official soon.

--bw