ImGui frame profiler overlay, and Vulkan per-draw descriptor set churn fix

[The-E]

This branch contains two related pieces of work: an in-game ImGui frame profiler
overlay, and a Vulkan descriptor-set optimisation that the overlay was used to
find and verify.

1. ImGui frame profiler overlay

Vendors ImPlot (lib/implot/, updated to the v1.1 WIP) and adds a runtime frame
profiler overlay on top of the existing tracing infrastructure:

• code/tracing/ProfilerOverlay.{h,cpp} — the overlay itself: a frametime history
  plot plus a breakdown of the biggest per-category contributors to the frame.
• FrameProfiler gains accumulate_self_times(), computing per-category
  exclusive (self) time for a frame's trace events in a single pass over the
  event stream, rather than building a tree and comparing strings. Split out so it
  can be unit-tested directly — see test/src/tracing/test_frame_profiler.cpp.
• Toggleable at runtime via the Game.ProfilerOverlay option; -profile_frame_time
  still works and now seeds the same toggle.
• Adds graphics/Vulkan debug counters (draw calls, descriptor sets and writes,
  pipeline count) to the overlay.

2. Vulkan: stop per-draw descriptor set churn

The Material and PerDraw descriptor sets were being reallocated and fully
rewritten on essentially every draw call. model_draw_list::render_buffer rebinds
ModelData with a fresh uniform buffer offset for each queued draw, and that
offset was part of the Material set's memoization key, so the cache in
VulkanDrawManager::applyMaterial missed every time — taking a frame-pool
allocation, a full template write, the 16-entry material texture resolution and a
vkUpdateDescriptorSets with it. GenericData/Matrices did the same to the
PerDraw set. renderShadowDraw was worse still: all three sets allocated and
rewritten per shadow draw, with no memoization at all.

The fix declares the bindings whose offset moves per draw
(MaterialBinding::ModelData, MaterialBinding::ShadowMapData,
PerDrawBinding::GenericData, PerDrawBinding::Matrices) as
eUniformBufferDynamic. DescriptorWriter::setBuffer splits a dynamic binding's
offset out of the descriptor — writing {buffer, 0, range} and stashing the offset
for vkCmdBindDescriptorSets' pDynamicOffsets — so an offset-only change leaves
the set's contents identical and the memoization caches start hitting.
Shader sources are unchanged.

VulkanStateTracker::bindDescriptorSet now compares the dynamic offsets as well as
the set handle. That part is required rather than an optimisation: the common case
is now the same set rebound with a new offset, and the old handle-only redundancy
check would have skipped that bind and fed every draw the previous draw's uniforms.

Also switches allocateFrameSet to the non-allocating allocateDescriptorSets
overload (the vector-returning one heap-allocated once per set, thousands of times
a frame). MAX_SETS_PER_POOL is deliberately left at 1024 — the per-frame pool
growth it used to log was a symptom of the churn, and one chunk now covers a dense
scene outright.

Measured

Solaris mission 21 (large asteroid field, ~1370 model draws/frame), RTX 4080 SUPER
at 1080p. Debug build, A/B on the same tree with and without the change:

	before	after
descriptor sets / frame	2566	250–345
descriptor writes / frame	19107	~1700–2300
median frametime (2 runs each)	92.9 / 83.6 ms	72.0 / 66.5 ms
Render Buffer (overlay)	24.68 ms (30.8%)	12.73 ms (18.4%)
Build Shadow Map (overlay)	5.44 ms	off the top-5 entirely
frame pool exhausted log lines	every frame	0

Release builds already inlined most of this per-draw work away and are unchanged
within run-to-run noise, which is the expected result — Render Buffer costs only
~1.4 µs/draw there. Worth noting: neither Release run happened to reach the heavy

600-draw stretch of the mission, so Release behaviour at very high draw counts is
untested; the draw-count bins that do overlap show no regression.

Testing

• Renders identically before/after at the same mission point (models, lighting,
  HUD, shadows) — no wrong-uniform corruption.
• Run under -gr_sync_validation -gr_debug: the set of reported VUIDs is identical
  to the same build without the change. No descriptor or dynamic-offset VUIDs.
• unittests 233/233 pass.
• Compiles warning-clean with -DFSO_FATAL_WARNINGS=ON (GCC).
• OpenGL is untouched by part 2.

[notimaginative]

This just renders a black screen for me on Mac. The one exception is that text renders normally, though only if the font is TTF, otherwise it too is black/missing.

No apparent errors or log info of note that I can see compared to a build from master that works fine.
fs2_open.log

I should also note that the lab appears to render normally. Until you load a model, at which point everything disappears except for the text.

[notimaginative]

It's working for me now, though I had to clear the cached shaders first. I'd tried that previously without success.

The [MoltenVK]: VK_ERROR_FEATURE_NOT_PRESENT: Metal does not support disabling primitive restart. message is gone now as well. Doesn't show up in the logs or on the console.