Reduce the importance sampling bias

Assets/ScriptableRenderLoop/HDRenderLoop/Sky/Resources/GGXConvolve.shader

             #endif
 
             float _Level;
+            float _MaxLevel;
             float _InvOmegaP;
 
             half4 Frag(Varyings input) : SV_Target
                     float4 val = IntegrateLD(TEXTURECUBE_PARAM(_MainTex, sampler_MainTex),
                                              V, N,
                                              roughness,
+                                             _MaxLevel,
                                              _InvOmegaP,
                                              sampleCount, // Must be a Fibonacci number
                                              true);

Assets/ScriptableRenderLoop/HDRenderLoop/Sky/SkyManager.cs

                 float invOmegaP = (6.0f * input.width * input.width) / (4.0f * Mathf.PI); // Solid angle associated to a pixel of the cubemap;
 
                 m_GGXConvolveMaterial.SetTexture("_MainTex", input);
+                m_GGXConvolveMaterial.SetFloat("_MaxLevel", mipCount - 1);
                 m_GGXConvolveMaterial.SetFloat("_InvOmegaP", invOmegaP);
 
                 for (int mip = 1; mip < ((int)EnvConstants.SpecCubeLodStep + 1); ++mip)

Assets/ScriptableRenderLoop/ShaderLibrary/ImageBasedLighting.hlsl

                     float3 V,
                     float3 N,
                     float roughness,
+                    float maxMipLevel,
                     float invOmegaP,
                     uint sampleCount, // Must be a Fibonacci number
                     bool prefilter)
         // Bias samples towards the mirror direction to reduce variance.
         // This will have a side effect of making the reflection sharper.
         // Ref: Stochastic Screen-Space Reflections, p. 67.
-        const float bias = 0.5;
+        const float bias = 0.2;
         u.x = lerp(u.x, 0.0, bias);
 
         float3 L;
 
             // Bias the MIP map level to compensate for the importance sampling bias.
             // This will blur the reflection.
-            // TODO: bias more accurately once the 'UNITY_SPECCUBE_LOD_STEPS' limit has been lifted.
-            mipLevel = lerp(mipLevel, UNITY_SPECCUBE_LOD_STEPS, sqrt(bias));
-            mipLevel = min(mipLevel, UNITY_SPECCUBE_LOD_STEPS);
+            mipLevel = lerp(mipLevel, maxMipLevel, bias);
         }
 
         if (NdotL > 0.0)