G-Buffer video memory compression

[japsuu]

Is your feature request related to a problem? Please describe.
Currently, the G-Buffer takes up 100 MB video memory.

Not all the current buffers are necessary, and some have unnecessarily large texture formats.

The following buffers are currently used:

Albedo & AO: RGBA16F = 8 bytes
Normal & Metalness: RGBA16F = 8 bytes
Position & Roughness: RGBA16F = 8 bytes
Emission: RGB16F = 6 bytes
Velocity: RG16F = 4 bytes
ObjectID: R32F = 4 bytes
Unlit: RGBA16F = 8 bytes
Depth: Depth24 = 3 bytes

Which is 49 bytes per pixel.

For a 1920x1080 screen that would make
1920x1080x49 = 101606400 bytes (101 MB)

Describe the solution you'd like
Remove the unnecessary buffers (unlit) and compress the existing ones further.
For example, normals could be stored using Signed Octahedron Normal Encoding

Additional context
https://www.reddit.com/r/opengl/comments/z2kdgm/deferred_rendering_reducing_the_size_of_the/
https://aras-p.info/texts/CompactNormalStorage.html
https://johnwhite3d.blogspot.com/2017/10/signed-octahedron-normal-encoding.html