improve performance of the graph aggregation (#8686)

### Description

Improves performance for larger applications by up to 30%
improves memory usage a bit

### Testing Instructions

<!--
  Give a quick description of steps to test your changes.
-->

crates/turbo-tasks-memory/src/aggregation/followers.rs

@@ -49,8 +49,8 @@ pub fn on_added<C: AggregationContext>(
     };
     let followers_len = aggregating.followers.len();
     let optimize = (!already_optimizing_for_node
-        && followers_len > MAX_FOLLOWERS
-        && (followers_len - MAX_FOLLOWERS).count_ones() == 1)
+        && followers_len >= MAX_FOLLOWERS
+        && followers_len.count_ones() == 1)
         .then(|| {
             aggregating
                 .followers

crates/turbo-tasks-memory/src/aggregation/increase.rs

@@ -4,7 +4,7 @@ use super::{
     balance_queue::BalanceQueue, AggegatingNode, AggregationContext, AggregationNode,
     AggregationNodeGuard, PreparedInternalOperation, PreparedOperation, StackVec,
 };
-pub(super) const LEAF_NUMBER: u32 = 64;
+pub(super) const LEAF_NUMBER: u32 = 16;
 
 #[derive(Debug)]
 pub enum IncreaseReason {
@@ -179,21 +179,23 @@ impl<C: AggregationContext> PreparedInternalOperation<C>
                 uppers,
                 reason,
             } => {
-                let mut need_to_run = true;
-                while need_to_run {
-                    need_to_run = false;
-                    let mut max = 0;
-                    for upper_id in &uppers {
-                        let upper = ctx.node(upper_id);
-                        let aggregation_number = upper.aggregation_number();
-                        if aggregation_number != u32::MAX {
-                            if aggregation_number > max {
-                                max = aggregation_number;
-                            }
-                            if aggregation_number == target_aggregation_number {
-                                target_aggregation_number += 1;
-                                if max >= target_aggregation_number {
-                                    need_to_run = true;
+                if target_aggregation_number >= LEAF_NUMBER {
+                    let mut need_to_run = true;
+                    while need_to_run {
+                        need_to_run = false;
+                        let mut max = 0;
+                        for upper_id in &uppers {
+                            let upper = ctx.node(upper_id);
+                            let aggregation_number = upper.aggregation_number();
+                            if aggregation_number != u32::MAX {
+                                if aggregation_number > max {
+                                    max = aggregation_number;
+                                }
+                                if aggregation_number == target_aggregation_number {
+                                    target_aggregation_number += 1;
+                                    if max >= target_aggregation_number {
+                                        need_to_run = true;
+                                    }
                                 }
                             }
                         }

crates/turbo-tasks-memory/src/aggregation/tests.rs

@@ -598,7 +598,7 @@ fn chain() {
         let aggregated = aggregation_data(&ctx, &current);
         assert_eq!(aggregated.value, 15050);
     }
-    assert_eq!(ctx.additions.load(Ordering::SeqCst), 122);
+    assert_eq!(ctx.additions.load(Ordering::SeqCst), 182);
     ctx.additions.store(0, Ordering::SeqCst);
     check_invariants(&ctx, once(current.clone()));
 
@@ -609,8 +609,8 @@ fn chain() {
     check_invariants(&ctx, once(current.clone()));
 
     leaf.incr(&ctx);
-    // The change need to propagate through 2 aggregated nodes
-    assert_eq!(ctx.additions.load(Ordering::SeqCst), 2);
+    // The change need to propagate through 4 aggregated nodes
+    assert_eq!(ctx.additions.load(Ordering::SeqCst), 4);
     ctx.additions.store(0, Ordering::SeqCst);
 
     {
@@ -640,7 +640,7 @@ fn chain() {
 
     leaf.incr(&ctx);
     // This should be less the 20 to prove that we are reusing trees
-    assert_eq!(ctx.additions.load(Ordering::SeqCst), 3);
+    assert_eq!(ctx.additions.load(Ordering::SeqCst), 5);
     ctx.additions.store(0, Ordering::SeqCst);
 
     {
@@ -678,10 +678,10 @@ fn chain_double_connected() {
 
     {
         let aggregated = aggregation_data(&ctx, &current);
-        assert_eq!(aggregated.value, 13188);
+        assert_eq!(aggregated.value, 20017);
     }
     check_invariants(&ctx, once(current.clone()));
-    assert_eq!(ctx.additions.load(Ordering::SeqCst), 285);
+    assert_eq!(ctx.additions.load(Ordering::SeqCst), 643);
     ctx.additions.store(0, Ordering::SeqCst);
 
     print(&ctx, &current, true);

crates/turbo-tasks-memory/src/aggregation/uppers.rs

@@ -97,10 +97,9 @@ pub fn on_added<C: AggregationContext>(
 ) -> usize {
     let uppers = node.uppers();
     let uppers_len = uppers.len();
-    let optimize = (!already_optimizing_for_upper
-        && uppers_len > MAX_UPPERS
-        && (uppers_len - MAX_UPPERS).count_ones() == 1)
-        .then(|| (true, uppers.iter().cloned().collect::<StackVec<_>>()));
+    let optimize =
+        (!already_optimizing_for_upper && uppers_len >= MAX_UPPERS && uppers_len.count_ones() == 1)
+            .then(|| (true, uppers.iter().cloned().collect::<StackVec<_>>()));
     let (add_change, followers) = match &mut *node {
         AggregationNode::Leaf { .. } => {
             let add_change = node.get_add_change();