Make batch_and_prepare_binned_render_phase only record information about the first batch in each batch set. (#17680)

Data for the other batches is only accessed by the GPU, not the CPU, so
it's a waste of time and memory to store information relating to those
other batches.

On Bistro, this reduces time spent in
`batch_and_prepare_binned_render_phase` from 85.9 us to 61.2 us, a 40%
speedup.

![Screenshot 2025-02-04
093315](https://github.com/user-attachments/assets/eb00db93-a260-44f9-9ae0-4e90b0697138)

Author: pcwalton

crates/bevy_render/src/batching/gpu_preprocessing.rs

@@ -1362,15 +1362,16 @@ pub fn batch_and_prepare_binned_render_phase<BPI, GFBD>(
                     match batch_set {
                         None => {
                             batch_set = Some(BinnedRenderPhaseBatchSet {
-                                batches: vec![batch],
+                                first_batch: batch,
+                                batch_count: 1,
                                 bin_key: bin_key.clone(),
                                 index: indirect_parameters_buffers
                                     .batch_set_count(batch_set_key.indexed())
                                     as u32,
                             });
                         }
                         Some(ref mut batch_set) => {
-                            batch_set.batches.push(batch);
+                            batch_set.batch_count += 1;
                         }
                     }
                 }
@@ -1498,7 +1499,8 @@ pub fn batch_and_prepare_binned_render_phase<BPI, GFBD>(
                         // However, custom render pipelines might do so, such as
                         // the `specialized_mesh_pipeline` example.
                         vec.push(BinnedRenderPhaseBatchSet {
-                            batches: vec![batch],
+                            first_batch: batch,
+                            batch_count: 1,
                             bin_key: key.1.clone(),
                             index: indirect_parameters_buffers.batch_set_count(key.0.indexed())
                                 as u32,

crates/bevy_render/src/render_phase/mod.rs

@@ -189,9 +189,16 @@ pub enum BinnedRenderPhaseBatchSets<BK> {
     MultidrawIndirect(Vec<BinnedRenderPhaseBatchSet<BK>>),
 }
 
+/// A group of entities that will be batched together into a single multi-draw
+/// call.
 pub struct BinnedRenderPhaseBatchSet<BK> {
-    pub(crate) batches: Vec<BinnedRenderPhaseBatch>,
+    /// The first batch in this batch set.
+    pub(crate) first_batch: BinnedRenderPhaseBatch,
+    /// The key of the bin that the first batch corresponds to.
     pub(crate) bin_key: BK,
+    /// The number of batches.
+    pub(crate) batch_count: u32,
+    /// The index of the batch set in the GPU buffer.
     pub(crate) index: u32,
 }
 
@@ -527,9 +534,7 @@ where
                     )
                     .zip(batch_sets.iter())
                 {
-                    let Some(batch) = batch_set.batches.first() else {
-                        continue;
-                    };
+                    let batch = &batch_set.first_batch;
 
                     let batch_set_index = if multi_draw_indirect_count_supported {
                         NonMaxU32::new(batch_set.index)
@@ -549,8 +554,7 @@ where
                             }
                             PhaseItemExtraIndex::IndirectParametersIndex { ref range, .. } => {
                                 PhaseItemExtraIndex::IndirectParametersIndex {
-                                    range: range.start
-                                        ..(range.start + batch_set.batches.len() as u32),
+                                    range: range.start..(range.start + batch_set.batch_count),
                                     batch_set_index,
                                 }
                             }