Possible `Locals::push` amortization improvement in `run_local_query` workload

[RoloEdits]

When doing some memory profiling, I noticed that Locals::push ends up requiring a lot grows:

The location for the initial pushes appears to be here:

tree-house/highlighter/src/locals.rs

Lines 209 to 217 in 16e1189

	scope = locals.push(ScopeData {
	definitions: HashMap::new(),
	range: matched_node.node.byte_range(),
	inherit: !injection_query
	.not_scope_inherits
	.contains(&query_match.pattern()),
	children: Vec::new(),
	parent: Some(scope),
	});

With push also calling the inner Vec::push:

tree-house/highlighter/src/locals.rs

Lines 44 to 52 in 16e1189

	fn push(&mut self, scope: ScopeData) -> Scope {
	let new_scope_id = Scope(self.scopes.len() as u32);
	let parent = scope
	.parent
	.expect("push cannot be used for the root layer");
	self[parent].children.push(new_scope_id);
	self.scopes.push(scope);
	new_scope_id
	}

locals is instantiated via Locals::default. Currently the default impl is using 4 as a starting point:

tree-house/highlighter/src/locals.rs

Lines 28 to 41 in 16e1189

	impl Default for Locals {
	fn default() -> Self {
	let mut scopes = Vec::with_capacity(4);
	scopes.push(ScopeData {
	definitions: HashMap::new(),
	range: 0..u32::MAX,
	inherit: false,
	children: Vec::new(),
	parent: None,
	});
	Self { scopes }
	}
	}

I wonder if increasing the starting size a bit could end up removing the need for some of the grows? Maybe try 8? Or 16? I don't know enough about the full context of how this function works to know if this is a meaningful change or not, but thought I would share.

As far as other places for large allocation numbers (not amount allocated, just the amount of allocations), most other areas in the screenshot lead directly into the C tree-sitter implementation. Of the spots in the actual Rust code, found just a few so far.