This commit adds support for copying nested hierarchical types of split
types.  E.g, a struct of a struct containing both user and builtin interstage
IO variables.

When copying split types, if any subtree does NOT contain builtin interstage
IO, we can copy the whole subtree with one assignment, which saves a bunch
of AST verbosity for memberwise copies of that subtree.

This adds structure splitting, which among other things will enable GS support where input structs
are passed, and thus become input arrays of structs in the GS inputs.  That is a common GS case.

The salient points of this PR are:

* Structure splitting has been changed from "always between stages" to "only into the VS and out of
  the PS".  It had previously happened between stages because it's not legal to pass a struct
  containing a builtin IO variable.

* Structs passed between stages are now split into a struct containing ONLY user types, and a
  collection of loose builtin IO variables, if any.  The user-part is passed as a normal struct
  between stages, which is valid SPIR-V now that the builtin IO is removed.

* Internal to the shader, a sanitized struct (with IO qualifiers removed) is used, so that e.g,
  functions can work unmodified.

* If a builtin IO such as Position occurs in an arrayed struct, for example as an input to a GS,
  the array reference is moved to the split-off loose variable, which is given the array dimension
  itself.

When passing things around inside the shader, such as over a function call, the the original type
is used in a sanitized form that removes the builtIn qualifications and makes them temporaries.
This means internal function calls do not have to change.  However, the type when returned from
the shader will be member-wise copied from the internal sanitized one to the external type.
The sanitized type is used in variable declarations.

When copying split types and unsplit, if a sub-struct contains only user variables, it is copied
as a single entity to avoid more AST verbosity.

Above strategy arrived at with talks with @johnkslang.

This is a big complex change.  I'm inclined to leave it as a WIP until it can get some exposure to
real world cases.

Modify shader ballot extension by adding OpSubgroupReadInvocationKHR

Add support for SPV_NV_geometry_shader_passthrough

Fix a large number of Clang warnings about inconsistent usage of 'override'

Also, eliminate the 'atom' field of TPpToken.

Parsing a real 300 line shader, through to making the AST, is about 10% faster.

Memory is slightly reduced (< 1%).

The whole google-test suite, inclusive of all testing overhead, SPIR-V generation,
etc., runs 3% faster.

Since this is a code *simplification* that leads to perf. improvement, I'm not
going to invest too much more in measuring the perf. than this. The PP code is
simply now in a better state to see how to further rationalize/improve it.

Always keep 'token' outside.
Always return the string to upper levels inside.

CMake: Add option to make it possible to not build the executables

Add support for SPV_NV_sample_mask_override_coverage

Removed the preprocesser memory pool.

Removed extra copies and unnecessary allocations of objects related to the ones
that were using the pool.

Replaced some allocated pointers with objects instead, generally using more
modern techiques. There end up being fewer memory allocations/deletions to get right.

Overall combined effect of all changes is to use slightly less memory and
run slightly faster (< 1% for both, but noticable).

As part of simplifying the code base, this change makes it easier to see
PP symbol tracking, which I suspect has an even bigger run-time simplification
to make.

Also, checking both 'atom' and 'name' is redundant, and I'm hoping to
eliminate more atom stuff.

PP: Implement token pasting for PP identifiers.

WIP: Remapper: remove debug info for IDs stripped in other passes

HLSL: allow "sample" in expressions.

Unbreak build on FreeBSD with GCC/libstdc++ 6.2.0

Implement token pasting as per the C++ specification, within the current
style of the PP code.
Non-identifiers (turning 12 ## 10 into the numeral 1210) is not yet covered;
they should be a simple incremental change built on this one.
Addresses issue #255.

Unlike other qualifiers, HLSL allows "sample" to be either a qualifier keyword or an
identifier (e.g, a variable or function name).

A fix to allow this was made a while ago, but that fix was insufficient when 'sample'
was used in an expression.  The problem was around the initial ambiguity between:

   sample float a; // "sample" is part of a fully specified type
and
   sample.xyz;     // sample is a keyword in a dot expression

Both start the same.  The "sample" was being accepted as a qualifier before enough
further parsing was done to determine we were not a declaration after all.  This
consumed the token, causing it to fail for its real purpose.

Now, when accepting a fully specified type, the token is pushed back onto the stack if
the thing is not a fully specified type.  This leaves it available for subsequent
purposes.

Changed the "hlsl.identifier.sample.frag" test to exercise this situation, distilled
down from a production shaders.

Change unicode dash to ASCII.

This change is helpful for integration with Chromium, which recently
added a compiler option to warn when compiling any source files which
use extended characters. In this case the offending character was a
single unicode dash in a comment.

Fix size_t to int cast warnings.

Several instances in Visual Studio 2015:

warning C4267: 'initializing': conversion from 'size_t' to 'int', possible loss of data

If some DCE is performed such as removing dead functions, then even
if we are NOT stripping debug info, we still must remove the debug
opcodes that refer to the now-dead IDs.

Also, this adds a small change to perform no ID remapping if none
is requested, making spirv-remap properly be a no-op if no options
are given.

Fixes issue #610. Also provides a testing option to keep uncalled functions.

This wasn't needed until the recent generalization of "main" to "entry point",
so makes some HLSL-specific code be generic now, for GLSL functional correctness.

In file included from C:/Projects/glslang/glslang/MachineIndependent/glslang.y:59:0:
glslang/MachineIndependent/ParseHelper.h:276:24: error: 'va_list' has not been declared
                        va_list args);
                        ^~~~~~~