MIS and delta lights

This is another minor issue with the new shading system. It looks as though MIS is broken, at least for delta lights.

This is rendering the default Vela scene; the sphere is glossy (roughness=0, which gets clamped to something small). I'm using Visionaray as reference for rendering with path tracing and NEE/MIS. I'm directing your attention towards the noise in the specular highlight from the directional light on the sphere surface.

Left is Barney.

1 SPP:

4 SPP:

64 SPP:

Note how in the reference images (with MIS to discriminate light and BRDF samples using the power heuristic) the highlight is virtually converged fully after one sample. This is because the light source is delta, the BRDF is nearly delta, and thus the light sampling PDF should be close to 1.0 and the BSDF PDF close to 0.0. I haven't checked the implementation at all, but it reminisces of an issue with how the MIS weights are applied.

Again adding @tarcila, as discussed earlier on Slack.

Reproduce the issue with the default Vela scene and the glossy sphere, comparing Barney's 1, 4, and 64 SPP output with the referenced Visionaray behavior. Trace the MIS weighting for directional or other delta lights and the BRDF sampling PDFs, then verify that the specular highlight converges with substantially less noise.