README links to "adjust input color" section, which was removed in 2.2

[Artoria2e5]

The README retains three references to the "adjust input color" stage which need to be removed:

• The L pyramid step still mentions AIC as a downstream step

  FidelityFX-FSR2/README.md

  Line 533 in 1680d1e

  The [Compute luminance pyramid](#compute-luminance-pyramid) stage is implemented using FidelityFX [Single Pass Downsampler](https://github.com/GPUOpen-Effects/FidelityFX-SPD), an optimized technique for producing mipmap chains using a single compute shader dispatch. Instead of the conventional (full) pyramidal approach, SPD provides a mechanism to produce a specific set of mipmap levels for an arbitrary input texture, as well as performing arbitrary calculations on that data as we store it to the target location in memory. In FSR2, we are interested in producing in upto two intermediate resources depending on the configuration of the [`FfxFsr2Context`](src/ffx-fsr2-api/ffx_fsr2.h#L166). The first resource is a low-resolution representation of the current luminance, this is used later in FSR2 to attempt to detect shading changes. The second is the exposure value, and while it is always computed, it is only used by subsequent stages if the [`FFX_FSR2_ENABLE_AUTO_EXPOSURE`](src/ffx-fsr2-api/ffx_fsr2.h#L93) flag is set in the [`flags`](src/ffx-fsr2-api/ffx_fsr2.h#L105) field of the [`FfxFsr2ContextDescription`](src/ffx-fsr2-api/ffx_fsr2.h#L103) structure upon context creation. The exposure value - either from the application, or the [Compute luminance pyramid](#compute-luminance-pyramid) stage - is used in the [Adjust input color](#adjust-input-color) stage of FSR2, as well as by the [Reproject & Accumulate](#project-and-accumulate) stage.

• Three explanations of R32_UNORM still mention AIC, e.g. https://github.com/GPUOpen-Effects/FidelityFX-FSR2/blob/1680d1edd5c034f88ebbbb793d8b88f8842cf804/README.md?plain=1#L589C662-L589C662

I am not sure about what whether the pyramid thing can be removed, or replaced with something new. The R32_UNORM explanation seems to only need this paragraph from e812d9e to be restored and placed in a more appropriate location:

The format of the previous depth buffer is R32_UINT which allows the use of InterlockedMax and InterlockedMin operations to be performed from the Reconstruct & dilate stage of FSR2. This is done with the resulting integer values returned by converting depth values using the asint functions. This works because depth values are always greater than 0, meaning that the monotonicity of IEEE754 floating point values when interpreted as integers is guaranteed.