Solari brdf fixes

[dylansechet]

Objective

BRDF fixes for solari!

Solution

• Use a local TBN instead of mikktspace, which was producing seams on the spheres used for the white furnace test.
• Separate metallic/dielectric paths, evaluating each endpoint separately then blending rather than working with a blended F0. This is the same as what Fix energy loss in multi-scattering term #23203 did for IBL.
• The Fresnel layering that was introduced isn't energy conserving. Instead, I compute the total energy reflected by the specular lobe using the DFG table and Filament's multi-scattering approximation. I can then scale the diffuse by the complement to ensure energy conservation.
  A downside to this approach is that it breaks reciprocity because of the simplifications that Filament's multi-scattering formulation makes. As far as I can tell the path tracer is unidirectional, so this shouldn't cause issues. The fix to restore reciprocity would be to add the full Kulla-Conty compensation lobe, which requires an extra LUT.

Note: I've mostly focused on getting this correct, performance can probably be improved. For example, this does entirely too many F_AB lookups.

Testing and showcase

White furnace

We're still not passing, but the changes improve it a lot.

On main:

With the LUT-based F_AB:

Adding the local TBN:

Adding the BRDF fixes:

Bunch of spheres

Notice the darkening on main for the middle row of spheres, with metallic=0.5.

Main:

This PR:

Dragons

The shading differences are particularly visible on the dragons in the front row.
There is some kind of exposure or tonemapping difference between bevy and cycles that's been driving me crazy, but I haven't been able to track it down.

Individual images

Main:

This PR:

Cycles:

More Dragons

Individual images

Main:

This PR:

Cycles:

[dylansechet]

Splitting into smaller PRs.