Clustered rendering can be done both deferred and forward.
They do a few things which are reminiscent of deferred, as you say. But what really makes a render deferred is the use of the gbuffer pass to only cull and collect data, followed by a rendering pass over a screen quad doing fragment shading with no potential overlap or wasted processing. Of course, you do get similar advantages from the depth prepass.
Isn't clustered forward rendering for lighting better in any aspect than "traditional" deferred shading, no matter if running on low- or high-end GPUs (because it wastes less memory bandwidth for accumulating light contribution from different light sources)? Also I think most rendering pipelines in modern games mix elements of both deferred and forward rendering (e.g. Doom Eternal renders additional information needed later in the frame into 'fat framebuffers', much like deferred renderers do).