drm: clear stale color state on modeset

[UncleJ4ck]

A previous compositor (kwin etc.) can leave HDR, gamma, degamma and CTM
properties set on a connector when it exits. Aquamarine only re-emitted
these on user commits, so anything we don't actively manage stayed at
the previous compositor's value, manifesting as washed-out colors or
shifted saturation when starting Hyprland after kwin.

HDR (#250) and CTM-with-non-identity (#256) were already covered. Extend
the modeset path to also:

• re-send gamma_lut / degamma_lut on modeset; an empty STATE.gammaLut
  produces a zero blob, which clears the kernel state. only ride this
  path when the prop actually exists, otherwise we'd log a spurious
  no gamma_lut prop error on every modeset for connectors without
  programmable gamma.
• re-send CTM unconditionally on modeset, identity included, so the
  kernel always replaces a stale matrix with our current one. the blob
  builder produces a real identity matrix when STATE.ctm == Mat3x3(),
  so we never send blob_id=0 (which drm: re-send ctm blob on modeset #256 documented as problematic).

Fixes #127.

[cushycush]

Hey, a heads up that I think it might've introduced a black-screen on some setups, wanted to flag it in case I'm reading the path wrong.

I'm running a Hyprland 0.55-based fork with render:cm_enabled=false on Intel UHD 770 / i915. Under atomic mode, every panel goes black after the first modeset. Legacy DRM works fine. The framebuffer itself is correct: grim reads back the rendered desktop. But modetest -M i915 -p shows the CRTC's CTM property is all-zeros after the first modeset, so the kernel multiplies every output pixel by zero.

The PR's commit message says the blob builder produces identity when STATE.ctm == Mat3x3(), but I don't think it does. Hyprutils::Math::Mat3x3() zero-initializes its 9 floats, and the builder at impl/Atomic.cpp:419-423 runs the matrix through doubleToS3132Fixed and writes it verbatim. So a default-constructed Mat3x3 lands in the kernel as a zero blob, not identity. Hyprland's setCTM only fires from the cm_enabled=true renderer path, so forks running with it disabled never overwrite the zero default.

Easy fix would be treating a default-constructed Mat3x3 as identity inside the blob builder, contained to this file. Flipping Hyprutils::Math::Mat3x3()'s default ctor to identity upstream would also close it but has broader blast radius for anyone quietly relying on zero. Anyone hitting this before an upstream fix can seed setCTM(identity()) at output creation as a workaround (no-op under legacy, seeds an identity matrix under atomic instead of all-zeros). Happy to send a PR if it helps, but figured I'd flag and let you take it since you're already in this file. Am I missing why the default ctor would produce identity?