Allows usages of the constructor to avoid an unnecessary copy.

In the process remove implementation of SUATOM.MIN and SUATOM.MAX as
these require a distinction between U32 and S32. These have to be
implemented with imageCompSwap loop.

* Implement SULD as float.
* Remove conditional declaration of GL_ARB_shader_viewport_layer_array.

This instruction writes to a memory buffer shared with threads within
the same work group. It is known as "shared" memory in GLSL.

Implement VOTE using Nvidia's intrinsics. Documentation about these can
be found here
https://developer.nvidia.com/reading-between-threads-shader-intrinsics

Instead of using portable ARB instructions I opted to use Nvidia
intrinsics because these are the closest we have to how Tegra X1
hardware renders.

To stub VOTE on non-Nvidia drivers (including nouveau) this commit
simulates a GPU with a warp size of one, returning what is meaningful
for the instruction being emulated:

* anyThreadNV(value) -> value
* allThreadsNV(value) -> value
* allThreadsEqualNV(value) -> true

ballotARB, also known as "uint64_t(activeThreadsNV())", emits

VOTE.ANY Rd, PT, PT;

on nouveau's compiler. This doesn't match exactly to Nvidia's code

VOTE.ALL Rd, PT, PT;

Which is emulated with activeThreadsNV() by this commit. In theory this
shouldn't really matter since .ANY, .ALL and .EQ affect the predicates
(set to PT on those cases) and not the registers.

This commit takes care of implementing the F16 Variants of the 
conversion instructions and makes sure conversions are done.

Hardware testing revealed that SSY and PBK push to a different stack,
allowing code like this:

        SSY label1;
        PBK label2;
        SYNC;
label1: PBK;
label2: EXIT;

Reflect std::shared_ptr nature of Node on initializers and remove
constant members in nodes.

Add some commentaries.

Analysis passes do not have a good reason to depend on shader_ir.h to
work on top of nodes. This splits node-related declarations to their own
file and leaves the IR in shader_ir.h

Instead of having a vector of unique_ptr stored in a vector and
returning star pointers to this, use shared_ptr. While changing
initialization code, move it to a separate file when possible.

This is a first step to allow code analysis and node generation beyond
the ShaderIR class.

This allows for forming comment nodes without making unnecessary copies
of the std::string instance.

e.g. previously:

Comment(fmt::format("Base address is c[0x{:x}][0x{:x}]",
        cbuf->GetIndex(), cbuf_offset));

Would result in a copy of the string being created, as CommentNode()
takes a std::string by value (a const ref passed to a value parameter
results in a copy).

Now, only one instance of the string is ever moved around. (fmt::format
returns a std::string, and since it's returned from a function by value,
this is a prvalue (which can be treated like an rvalue), so it's moved
into Comment's string parameter), we then move it into the CommentNode
constructor, which then moves the string into its member variable).

constexpr internally links by default, so the inline specifier is
unnecessary.

Many of these constructors don't even need to be templated. The only
ones that need to be templated are the ones that actually make use of
the parameter pack.

Even then, since std::vector accepts an initializer list, we can supply
the parameter pack directly to it instead of creating our own copy of
the list, then copying it again into the std::vector.

shader/shader_ir: Remove unnecessary template parameter packs from Operation() overloads where applicable

These overloads don't actually make use of the parameter pack, so they
can be turned into regular non-template function overloads.

These don't actually modify instance state, so they can be marked as
const member functions

Given the class contains quite a lot of non-trivial types, place the
constructor and destructor within the cpp file to avoid inlining
construction and destruction code everywhere the class is used.

Operations done before the main half float operation (like HAdd) were
managing a packed value instead of the unpacked one. Adding an unpacked
operation allows us to drop the per-operand MetaHalfArithmetic entry,
simplifying the code overall.