bevy_pbr: indexed/non_indexed occlusion-culling bind-group fns differ only by buffer

While reading `crates/bevy_pbr/src/render/gpu_preprocess.rs` I noticed the occlusion-culling bind-group constructors pair up almost exactly along the indexed/non_indexed axis:

- `create_indirect_occlusion_culling_early_indexed_bind_group` (line 2090) and `create_indirect_occlusion_culling_early_non_indexed_bind_group` (line 2192) differ only in the `indexed` vs `non_indexed` buffer and field names; the bodies are otherwise identical (~91 lines each).
- The same holds for the `late_indexed` (line 2294) and `late_non_indexed` (line 2385) pair.

The early/late axis is genuinely different (early binds an extra work-item buffer that late does not), so that part is not duplication. But the indexed/non_indexed split looks like it could collapse into two functions parameterized over the buffer set, removing roughly half of the ~250 near-duplicate lines.

Is the explicit form deliberate (readability of the GPU bind-group setup), or would a PR parameterizing the indexed/non_indexed pairs be welcome? Happy to send one if so.

Read crates/bevy_pbr/src/render/gpu_preprocess.rs and compare the four named occlusion-culling bind-group constructors, starting with the indexed/non_indexed pairs. Confirm which buffer and field differences can be parameterized while preserving the early/late distinction. Done means the duplicated indexed/non-indexed setup is reduced without changing bind-group behavior.

Written by the indexing model from the issue text.